feat: More verbose logging on workflow logs

agent/internal/exec/exec.go

@@ -35,10 +35,21 @@ func (w *streamWriter) Write(p []byte) (int, error) {
 	return len(p), nil
 }
 
+// WorkspacePath returns the per-run working directory for a workspace id. The
+// same id always maps to the same path so RunStep and the cleanup command agree.
+func WorkspacePath(workspaceID string) string {
+	return filepath.Join(os.TempDir(), "vantage-run-"+workspaceID)
+}
+
 // RunStep writes the script to a temp file, provides a WORKFLOW_ENV file for
 // the script to append KEY=value output to, executes it under the requested
 // interpreter, and streams output via emit, returning the terminal result
 // with empty stdout/stderr but populated exit_code/output_env.
+//
+// When the command carries a WorkspaceId the step runs with that per-run working
+// directory as its cwd (created here if missing); the server removes it once the
+// run finishes. The script and env files always live in a private temp dir so
+// they never leak into the shared workspace.
 func RunStep(cmd *pb.RunStepCmd, emit func(seq uint64, data []byte)) *pb.StepResult {
 	res := &pb.StepResult{CommandId: "", OutputEnv: map[string]string{}}
 
@@ -50,6 +61,16 @@ func RunStep(cmd *pb.RunStepCmd, emit func(seq uint64, data []byte)) *pb.StepRes
 	}
 	defer os.RemoveAll(dir)
 
+	workDir := ""
+	if cmd.WorkspaceId != "" {
+		workDir = WorkspacePath(cmd.WorkspaceId)
+		if err := os.MkdirAll(workDir, 0700); err != nil {
+			res.ExitCode = 1
+			res.Stderr = "create workspace: " + err.Error()
+			return res
+		}
+	}
+
 	envFile := filepath.Join(dir, "workflow_env")
 	if err := os.WriteFile(envFile, nil, 0600); err != nil {
 		res.ExitCode = 1
@@ -91,6 +112,10 @@ func RunStep(cmd *pb.RunStepCmd, emit func(seq uint64, data []byte)) *pb.StepRes
 		c = exec.CommandContext(ctx, "bash", scriptPath)
 	}
 
+	if workDir != "" {
+		c.Dir = workDir
+	}
+
 	c.Env = append(os.Environ(), "WORKFLOW_ENV="+envFile)
 	for k, v := range cmd.Env {
 		c.Env = append(c.Env, k+"="+v)

agent/internal/grpc/pb/vantage.pb.go

@@ -63,12 +63,19 @@ type ReportUpdatesResponse struct{}
 type ApplyUpdatesCmd struct{}
 
 type ServerCommand struct {
-	CommandId    string           `json:"command_id"`
-	GenerateKey  *GenerateKeyCmd  `json:"generate_key,omitempty"`
-	DeleteKey    *DeleteKeyCmd    `json:"delete_key,omitempty"`
-	UpdateAgent  *UpdateAgentCmd  `json:"update_agent,omitempty"`
-	ApplyUpdates *ApplyUpdatesCmd `json:"apply_updates,omitempty"`
-	RunStep      *RunStepCmd      `json:"run_step,omitempty"`
+	CommandId        string               `json:"command_id"`
+	GenerateKey      *GenerateKeyCmd      `json:"generate_key,omitempty"`
+	DeleteKey        *DeleteKeyCmd        `json:"delete_key,omitempty"`
+	UpdateAgent      *UpdateAgentCmd      `json:"update_agent,omitempty"`
+	ApplyUpdates     *ApplyUpdatesCmd     `json:"apply_updates,omitempty"`
+	RunStep          *RunStepCmd          `json:"run_step,omitempty"`
+	CleanupWorkspace *CleanupWorkspaceCmd `json:"cleanup_workspace,omitempty"`
+}
+
+// CleanupWorkspaceCmd tells the agent to recursively remove the run's working
+// directory once all steps on that server have finished.
+type CleanupWorkspaceCmd struct {
+	WorkspaceId string `json:"workspace_id"`
 }
 
 type DeleteKeyCmd struct {
@@ -110,6 +117,9 @@ type RunStepCmd struct {
 	Script         string            `json:"script"`
 	Env            map[string]string `json:"env,omitempty"`
 	TimeoutSeconds int               `json:"timeout_seconds,omitempty"`
+	// WorkspaceId names the per-run working directory the agent creates and uses
+	// as the step's cwd. Empty means run in the agent's default directory.
+	WorkspaceId string `json:"workspace_id,omitempty"`
 }
 
 type StepResult struct {

agent/internal/sync/sync.go

@@ -198,6 +198,9 @@ func connectAndHandleStream(ctx context.Context, cfg *config.Config) error {
 		if cmd.ApplyUpdates != nil {
 			go handleApplyUpdates(cfg, cmd)
 		}
+		if cmd.CleanupWorkspace != nil {
+			go handleCleanupWorkspace(cmd)
+		}
 		if cmd.RunStep != nil {
 			go func(rc *pb.RunStepCmd, cid string) {
 				emit := func(seq uint64, data []byte) {
@@ -286,6 +289,16 @@ func handleApplyUpdates(cfg *config.Config, cmd *pb.ServerCommand) {
 	_ = client.ReportUpdates(cfg.ServerID, cfg.AgentToken, nil)
 }
 
+func handleCleanupWorkspace(cmd *pb.ServerCommand) {
+	id := cmd.CleanupWorkspace.WorkspaceId
+	dir := agentexec.WorkspacePath(id)
+	if err := os.RemoveAll(dir); err != nil {
+		log.Printf("cleanup workspace %s failed (cmd=%s): %v", dir, cmd.CommandId, err)
+		return
+	}
+	log.Printf("removed run workspace %s (cmd=%s)", dir, cmd.CommandId)
+}
+
 func handleDeleteKey(cmd *pb.ServerCommand) {
 	label := cmd.DeleteKey.Label
 	keyPath := fmt.Sprintf("/root/.ssh/vantage_%s", strings.ReplaceAll(label, " ", "_"))

docs/superpowers/plans/2026-07-20-workflow-log-streaming.md

@@ -1,887 +0,0 @@
-# Workflow Log Streaming Implementation Plan
-
-> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.
-
-**Goal:** Stream workflow step output live from agents to per-server-run log files on the server, tail them live in the UI over SSE, and auto-expire them on a configurable retention period.
-
-**Architecture:** Agent streams interleaved stdout/stderr chunks over the existing `CommandStream` (`AgentMessage.StepOutput`). Server appends secret-masked chunks to `<logdir>/<run_id>/<server_id>.log` via a per-command log-writer registry, records a per-step byte offset, and stops persisting log bodies in Mongo. UI tails via an SSE endpoint while running and fetches the whole file after. An hourly sweeper deletes run-log dirs older than the retention setting.
-
-**Tech Stack:** Go (gin, mongo-driver v2), hand-written JSON-codec gRPC structs (no protoc), Next.js 16 app-router + react-query + EventSource, MongoDB, local filesystem for logs.
-
-## Global Constraints
-
-- **No tests this iteration** — do not write `*_test.go` or frontend tests. Verify each task with `go build ./...`, `go vet ./...`, and (frontend) `npm run build`.
-- gRPC uses a **JSON codec** — proto messages are hand-written Go structs in **two** files that must stay identical: `server/internal/grpc/pb/vantage.pb.go` and `agent/internal/grpc/pb/vantage.pb.go`. No codegen. Also update `proto/vantage/v1/vantage.proto` as documentation.
-- Mongo access pattern: `db.Col("collection_name")` with `context.WithTimeout`. Follow `server/internal/services/workflows.go`.
-- Secret values must never be written into log files unmasked — mask by literal `***` replacement at write time, boundary-safe via a carry buffer.
-- Interpreter values are the literals `"bash"` and `"powershell"`.
-- Go module path: `github.com/mrhid6/vantage`.
-- Log dir from env `VANTAGE_WORKFLOW_LOG_DIR`, default `<data>/workflow-logs`; files `0600`, dirs `0700`.
-- Retention default **30** days, stored `settings.workflow_log_retention_days`; `0`/negative = keep forever.
-- The agent's stream `Send` is only safe through the existing per-connection mutex-guarded `send()` closure in `connectAndHandleStream` — all `StepOutput`/`StepResult` sends MUST go through it.
-
----
-
-## Task 1: Proto/pb — StepOutputChunk
-
-**Files:**
-- Modify: `proto/vantage/v1/vantage.proto`
-- Modify: `server/internal/grpc/pb/vantage.pb.go`
-- Modify: `agent/internal/grpc/pb/vantage.pb.go`
-
-**Interfaces:**
-- Produces: `pb.StepOutputChunk{CommandId string, Seq uint64, Data []byte, Eof bool}`; `pb.AgentMessage` gains `StepOutput *StepOutputChunk`.
-
-- [ ] **Step 1: Document in the proto file**
-
-In `proto/vantage/v1/vantage.proto`, add to the `AgentMessage` oneof: `StepOutputChunk step_output = 6;` and add the message:
-
-```protobuf
-message StepOutputChunk {
-  string command_id = 1;
-  uint64 seq        = 2;
-  bytes  data       = 3;
-  bool   eof        = 4;
-}
-```
-
-- [ ] **Step 2: Add struct + field to server pb file**
-
-In `server/internal/grpc/pb/vantage.pb.go`, add to `type AgentMessage struct { ... }`:
-
-```go
-	StepOutput *StepOutputChunk `json:"step_output,omitempty"`
-```
-
-and add the new struct:
-
-```go
-type StepOutputChunk struct {
-	CommandId string `json:"command_id"`
-	Seq       uint64 `json:"seq"`
-	Data      []byte `json:"data,omitempty"`
-	Eof       bool   `json:"eof,omitempty"`
-}
-```
-
-- [ ] **Step 3: Mirror identical additions into the agent pb file**
-
-Apply the identical `AgentMessage.StepOutput` field and `StepOutputChunk` struct to `agent/internal/grpc/pb/vantage.pb.go`.
-
-- [ ] **Step 4: Verify build**
-
-Run: `cd server && go build ./... && cd ../agent && go build ./...`
-Expected: both succeed.
-
-- [ ] **Step 5: Commit**
-
-```bash
-git add proto/vantage/v1/vantage.proto server/internal/grpc/pb/vantage.pb.go agent/internal/grpc/pb/vantage.pb.go
-git commit -m "feat(proto): add StepOutputChunk streaming message"
-```
-
----
-
-## Task 2: Agent — stream step output
-
-**Files:**
-- Modify: `agent/internal/exec/exec.go`
-- Modify: `agent/internal/sync/sync.go` (the `cmd.RunStep != nil` goroutine)
-
-**Interfaces:**
-- Consumes: `pb.RunStepCmd`, `pb.StepResult`, `pb.StepOutputChunk` (Task 1).
-- Produces: `exec.RunStep(cmd *pb.RunStepCmd, emit func(seq uint64, data []byte)) *pb.StepResult` — streams output via `emit`, returns terminal result with empty stdout/stderr but populated exit_code/output_env.
-
-- [ ] **Step 1: Rework `exec.RunStep` to stream**
-
-In `agent/internal/exec/exec.go`, change the signature and replace the two `bytes.Buffer`s with a single mutex-guarded streaming writer. Full new body of the run/capture section (keep the existing temp-dir, env-file, interpreter-selection, timeout, and `parseEnvFile` logic exactly as-is):
-
-Add this type at package scope:
-
-```go
-// streamWriter forwards every write to emit() as an ordered chunk. Used as both
-// Stdout and Stderr so output interleaves in real execution order. The mutex
-// ensures a single stdout/stderr write is not interleaved mid-slice with another.
-type streamWriter struct {
-	mu   sync.Mutex
-	seq  uint64
-	emit func(seq uint64, data []byte)
-}
-
-func (w *streamWriter) Write(p []byte) (int, error) {
-	w.mu.Lock()
-	defer w.mu.Unlock()
-	if w.emit != nil {
-		buf := make([]byte, len(p))
-		copy(buf, p)
-		w.emit(w.seq, buf)
-		w.seq++
-	}
-	return len(p), nil
-}
-```
-
-Add `"sync"` to the imports. Change the signature to:
-
-```go
-func RunStep(cmd *pb.RunStepCmd, emit func(seq uint64, data []byte)) *pb.StepResult {
-```
-
-Replace the block that currently declares `var stdout, stderr bytes.Buffer`, assigns `c.Stdout`/`c.Stderr`, and sets `res.Stdout`/`res.Stderr` from them, with:
-
-```go
-	sw := &streamWriter{emit: emit}
-	c.Stdout = sw
-	c.Stderr = sw
-	runErr := c.Run()
-
-	// stdout/stderr are streamed via emit, not returned in the result.
-	if ctx.Err() == context.DeadlineExceeded {
-		res.ExitCode = 124
-		res.Stderr = "[vantage] step timed out"
-	} else if ee, ok := runErr.(*exec.ExitError); ok {
-		res.ExitCode = ee.ExitCode()
-	} else if runErr != nil {
-		res.ExitCode = 1
-		res.Stderr = "[vantage] " + runErr.Error()
-	}
-
-	res.OutputEnv = parseEnvFile(envFile)
-	return res
-```
-
-Remove the now-unused `"bytes"` and `"bufio"` imports **only if** they are no longer referenced (`parseEnvFile` uses `bufio` + `os` — keep `bufio`; `bytes` is likely now unused — remove it if so). Verify with `go build`.
-
-- [ ] **Step 2: Wire streaming into the agent loop**
-
-In `agent/internal/sync/sync.go`, the `cmd.RunStep != nil` goroutine currently calls `agentexec.RunStep(rc)` and sends one `StepResult` via `send()`. Change it to pass an `emit` closure that streams chunks, then send an eof chunk, then the terminal result — all through the existing mutex-guarded `send()`:
-
-```go
-		if cmd.RunStep != nil {
-			go func(rc *pb.RunStepCmd, cid string) {
-				emit := func(seq uint64, data []byte) {
-					_ = send(&pb.AgentMessage{
-						ServerId:   cfg.ServerID,
-						AgentToken: cfg.AgentToken,
-						StepOutput: &pb.StepOutputChunk{CommandId: cid, Seq: seq, Data: data},
-					})
-				}
-				res := agentexec.RunStep(rc, emit)
-				res.CommandId = cid
-				// Final eof marker so the server closes the log file.
-				_ = send(&pb.AgentMessage{
-					ServerId:   cfg.ServerID,
-					AgentToken: cfg.AgentToken,
-					StepOutput: &pb.StepOutputChunk{CommandId: cid, Eof: true},
-				})
-				_ = send(&pb.AgentMessage{
-					ServerId:   cfg.ServerID,
-					AgentToken: cfg.AgentToken,
-					StepResult: res,
-				})
-			}(cmd.RunStep, cmd.CommandId)
-			continue
-		}
-```
-
-(Match the exact field names already used by the existing `send()` calls in this function — `cfg.ServerID`, `cfg.AgentToken`, and the `send` closure. If the existing RunStep branch used different local names, keep those.)
-
-- [ ] **Step 3: Verify build**
-
-Run: `cd agent && go build ./... && go vet ./...`
-Expected: success. Resolve any leftover unused-import error from Step 1.
-
-- [ ] **Step 4: Commit**
-
-```bash
-git add agent/internal/exec/exec.go agent/internal/sync/sync.go
-git commit -m "feat(agent): stream step output chunks over CommandStream"
-```
-
----
-
-## Task 3: Server log-writer registry + retention sweeper
-
-**Files:**
-- Create: `server/internal/services/steplogs.go`
-
-**Interfaces:**
-- Consumes: `settings` service (retention), `db.Col("workflow_runs")` (sweeper), env `VANTAGE_WORKFLOW_LOG_DIR`.
-- Produces:
-  - `WorkflowLogDir() string` — resolved base dir (env or default), created on first call.
-  - `ServerRunLogPath(runID, serverID string) string` — `<logdir>/<runID>/<serverID>.log`.
-  - `AppendMarker(runID, serverID, line string) (int64, error)` — appends a marker line, returns the byte offset **before** the write (the step's `log_offset`).
-  - `var StepLogs *stepLogRegistry` with `Open(commandID, path string, secrets []string) error`, `Append(commandID string, data []byte)`, `Close(commandID string)`.
-  - `StartLogSweeper()` — launches the hourly retention goroutine; also sweeps once immediately.
-
-- [ ] **Step 1: Write the registry, paths, and sweeper**
-
-```go
-package services
-
-import (
-	"os"
-	"path/filepath"
-	"strings"
-	"sync"
-	"time"
-
-	"github.com/mrhid6/vantage/server/internal/db"
-	"go.mongodb.org/mongo-driver/v2/bson"
-)
-
-// WorkflowLogDir returns the base directory for workflow step logs, creating it.
-func WorkflowLogDir() string {
-	dir := os.Getenv("VANTAGE_WORKFLOW_LOG_DIR")
-	if dir == "" {
-		dir = filepath.Join("data", "workflow-logs")
-	}
-	_ = os.MkdirAll(dir, 0700)
-	return dir
-}
-
-// ServerRunLogPath is the per-server-run log file path.
-func ServerRunLogPath(runID, serverID string) string {
-	return filepath.Join(WorkflowLogDir(), runID, serverID+".log")
-}
-
-// AppendMarker appends a line to the server-run log and returns the byte offset
-// at which the write began (used as a step's log_offset).
-func AppendMarker(runID, serverID, line string) (int64, error) {
-	path := ServerRunLogPath(runID, serverID)
-	if err := os.MkdirAll(filepath.Dir(path), 0700); err != nil {
-		return 0, err
-	}
-	f, err := os.OpenFile(path, os.O_APPEND|os.O_CREATE|os.O_WRONLY, 0600)
-	if err != nil {
-		return 0, err
-	}
-	defer f.Close()
-	off, _ := f.Seek(0, 2) // current end = offset before write
-	if _, err := f.WriteString(line); err != nil {
-		return off, err
-	}
-	return off, nil
-}
-
-// ---- streamed chunk writer, boundary-safe secret masking ----
-
-type stepLogWriter struct {
-	mu        sync.Mutex
-	f         *os.File
-	carry     []byte
-	secrets   []string
-	maxSecret int
-}
-
-type stepLogRegistry struct {
-	mu      sync.Mutex
-	writers map[string]*stepLogWriter
-}
-
-var StepLogs = &stepLogRegistry{writers: make(map[string]*stepLogWriter)}
-
-// Open opens (append) the server-run file for a step's streamed chunks.
-func (r *stepLogRegistry) Open(commandID, path string, secrets []string) error {
-	if err := os.MkdirAll(filepath.Dir(path), 0700); err != nil {
-		return err
-	}
-	f, err := os.OpenFile(path, os.O_APPEND|os.O_CREATE|os.O_WRONLY, 0600)
-	if err != nil {
-		return err
-	}
-	max := 0
-	for _, s := range secrets {
-		if len(s) > max {
-			max = len(s)
-		}
-	}
-	w := &stepLogWriter{f: f, secrets: secrets, maxSecret: max}
-	r.mu.Lock()
-	r.writers[commandID] = w
-	r.mu.Unlock()
-	return nil
-}
-
-func (r *stepLogRegistry) get(commandID string) *stepLogWriter {
-	r.mu.Lock()
-	defer r.mu.Unlock()
-	return r.writers[commandID]
-}
-
-// Append masks and writes a chunk, holding back the last maxSecret-1 bytes so a
-// secret split across a chunk boundary is still masked on the next append/close.
-func (r *stepLogRegistry) Append(commandID string, data []byte) {
-	w := r.get(commandID)
-	if w == nil {
-		return
-	}
-	w.mu.Lock()
-	defer w.mu.Unlock()
-	if len(w.secrets) == 0 || w.maxSecret <= 1 {
-		_, _ = w.f.Write(data)
-		return
-	}
-	buf := append(w.carry, data...)
-	hold := w.maxSecret - 1
-	if len(buf) <= hold {
-		w.carry = buf
-		return
-	}
-	flush := buf[:len(buf)-hold]
-	w.carry = append([]byte{}, buf[len(buf)-hold:]...)
-	_, _ = w.f.Write(maskBytes(flush, w.secrets))
-}
-
-// Close flushes the carry (masked) and closes the file.
-func (r *stepLogRegistry) Close(commandID string) {
-	r.mu.Lock()
-	w := r.writers[commandID]
-	delete(r.writers, commandID)
-	r.mu.Unlock()
-	if w == nil {
-		return
-	}
-	w.mu.Lock()
-	defer w.mu.Unlock()
-	if len(w.carry) > 0 {
-		_, _ = w.f.Write(maskBytes(w.carry, w.secrets))
-		w.carry = nil
-	}
-	_ = w.f.Close()
-}
-
-func maskBytes(b []byte, secrets []string) []byte {
-	s := string(b)
-	for _, v := range secrets {
-		if v == "" {
-			continue
-		}
-		s = strings.ReplaceAll(s, v, "***")
-	}
-	return []byte(s)
-}
-
-// ---- retention sweeper ----
-
-// StartLogSweeper sweeps expired run-log dirs hourly (and once now).
-func StartLogSweeper() {
-	go func() {
-		sweepLogs()
-		t := time.NewTicker(time.Hour)
-		defer t.Stop()
-		for range t.C {
-			sweepLogs()
-		}
-	}()
-}
-
-func sweepLogs() {
-	days := retentionDays()
-	if days <= 0 {
-		return
-	}
-	cutoff := time.Now().AddDate(0, 0, -days)
-	base := WorkflowLogDir()
-	entries, err := os.ReadDir(base)
-	if err != nil {
-		return
-	}
-	for _, e := range entries {
-		if !e.IsDir() {
-			continue
-		}
-		runID := e.Name()
-		dir := filepath.Join(base, runID)
-		if runExpired(runID, dir, cutoff) {
-			_ = os.RemoveAll(dir)
-		}
-	}
-}
-
-// runExpired is true when the run finished before cutoff (falling back to dir
-// mtime when the run doc is gone).
-func runExpired(runID, dir string, cutoff time.Time) bool {
-	ctx, cancel := wfCtx()
-	defer cancel()
-	var run struct {
-		FinishedAt *time.Time `bson:"finished_at"`
-	}
-	err := db.Col("workflow_runs").FindOne(ctx, bson.M{"run_id": runID}).Decode(&run)
-	if err == nil {
-		if run.FinishedAt == nil {
-			return false // still running / never finished — keep
-		}
-		return run.FinishedAt.Before(cutoff)
-	}
-	// run doc gone: use dir mtime
-	if fi, e := os.Stat(dir); e == nil {
-		return fi.ModTime().Before(cutoff)
-	}
-	return false
-}
-
-func retentionDays() int {
-	if v, err := GetWorkflowLogRetentionDays(); err == nil {
-		return v
-	}
-	return 30
-}
-```
-
-Note: `wfCtx` is defined in `workflows.go` (same package) — reuse it. `GetWorkflowLogRetentionDays` is added in Task 4; this file references it (same package, compiles together).
-
-- [ ] **Step 2: Verify build**
-
-Run: `cd server && go build ./... && go vet ./...`
-Expected: FAIL — `GetWorkflowLogRetentionDays` undefined until Task 4. This is expected; proceed to commit the file so Task 4 completes it. (If you prefer a green build, do Task 4's settings accessor first, then return — but committing here is fine since Task 4 immediately follows.)
-
-Actually to keep every commit buildable: **temporarily** add a local stub at the bottom of this file and remove it in Task 4:
-
-```go
-// TEMP stub, replaced in Task 4.
-func GetWorkflowLogRetentionDays() (int, error) { return 30, nil }
-```
-
-Then `cd server && go build ./... && go vet ./...` must succeed.
-
-- [ ] **Step 3: Commit**
-
-```bash
-git add server/internal/services/steplogs.go
-git commit -m "feat(server): workflow log-writer registry, paths, retention sweeper"
-```
-
----
-
-## Task 4: Settings — retention accessor + startup wiring
-
-**Files:**
-- Modify: `server/internal/services/settings.go` (or wherever settings get/set lives — search `settings` collection usage)
-- Modify: `server/internal/services/steplogs.go` (remove the temp stub)
-- Modify: `server/cmd/main.go` (start the sweeper)
-
-**Interfaces:**
-- Produces: `GetWorkflowLogRetentionDays() (int, error)` (default 30 when unset), `SetWorkflowLogRetentionDays(int) error`. If settings are exposed as a single document/struct, add the field there and derive these accessors.
-
-- [ ] **Step 1: Inspect the settings service**
-
-Read the existing settings service (search for the `settings` collection: `grep -rn "\"settings\"" server/internal/services`). Determine whether settings are a typed struct document or key/value. Match that pattern.
-
-- [ ] **Step 2: Add the retention accessor**
-
-If settings are a **typed document** (e.g. a `GetSettings()/UpdateSettings()`), add a field `WorkflowLogRetentionDays int `bson:"workflow_log_retention_days" json:"workflow_log_retention_days"`` to the settings struct and implement:
-
-```go
-func GetWorkflowLogRetentionDays() (int, error) {
-	s, err := GetSettings() // use the real accessor name
-	if err != nil {
-		return 30, err
-	}
-	if s.WorkflowLogRetentionDays == 0 && /* unset sentinel */ !s.WorkflowLogRetentionSet {
-		return 30, nil
-	}
-	return s.WorkflowLogRetentionDays, nil
-}
-```
-
-Simplify to match reality: if the settings doc uses zero-value-means-unset and you cannot distinguish "0 = keep forever" from "unset", store the retention as a pointer `*int` or default at read: **treat a missing field as 30, an explicit 0 as keep-forever.** Prefer `*int` in the struct so the three states (unset→30, 0→forever, N→N) are representable. Implement `GetWorkflowLogRetentionDays` to return 30 when the pointer is nil, else its value. `SetWorkflowLogRetentionDays(n int)` sets the pointer.
-
-If settings are **key/value**, implement both accessors against that store with the same nil→30 / 0→forever semantics (store empty/absent = 30).
-
-- [ ] **Step 3: Remove the temp stub from `steplogs.go`**
-
-Delete the `// TEMP stub` `GetWorkflowLogRetentionDays` added in Task 3 so the real one is used.
-
-- [ ] **Step 4: Start the sweeper at boot**
-
-In `server/cmd/main.go`, next to `EnsureWorkflowIndexes()`, add `services.StartLogSweeper()`.
-
-- [ ] **Step 5: Verify build**
-
-Run: `cd server && go build ./... && go vet ./...`
-Expected: success (real accessor now resolves the reference from Task 3).
-
-- [ ] **Step 6: Commit**
-
-```bash
-git add server/internal/services/settings.go server/internal/services/steplogs.go server/cmd/main.go
-git commit -m "feat(server): workflow log retention setting + sweeper startup"
-```
-
----
-
-## Task 5: Runner + model — write to files, drop log bodies from Mongo
-
-**Files:**
-- Modify: `server/internal/models/workflow.go` (`StepRun`)
-- Modify: `server/internal/services/workflow_runner.go`
-- Modify: `server/internal/grpc/server.go` (stream delivery of `StepOutput`)
-
-**Interfaces:**
-- Consumes: `StepLogs`, `AppendMarker`, `ServerRunLogPath` (Task 3), `pb.StepOutputChunk` (Task 1).
-- Produces: runner writes markers + streams chunks to files; `StepRun.LogOffset` persisted; `StepRun.Stdout/Stderr` removed.
-
-- [ ] **Step 1: Update the `StepRun` model**
-
-In `server/internal/models/workflow.go`, in `type StepRun struct`:
-- Remove the `Stdout` and `Stderr` fields.
-- Add: `LogOffset int64 `bson:"log_offset" json:"log_offset"``
-
-- [ ] **Step 2: Deliver StepOutput chunks in the gRPC receive loop**
-
-In `server/internal/grpc/server.go`, after the existing `if m.StepResult != nil { services.StepResults.Deliver(m.StepResult) }` block, add:
-
-```go
-			if m.StepOutput != nil {
-				if m.StepOutput.Eof {
-					services.StepLogs.Close(m.StepOutput.CommandId)
-				} else {
-					services.StepLogs.Append(m.StepOutput.CommandId, m.StepOutput.Data)
-				}
-			}
-```
-
-- [ ] **Step 3: Rework `runServer` to open logs + write markers, drop persisted bodies**
-
-In `server/internal/services/workflow_runner.go`, `runServer`:
-
-Inside the per-step loop, **before** `dispatchAndWait`, add marker + open (compute `secretVals` first, which already exists in the loop):
-
-```go
-		// Write the step marker and remember the offset for later slicing.
-		marker := fmt.Sprintf("\n===== step %d: %s =====\n", step.Order, step.Name)
-		offset, _ := AppendMarker(runID, serverID, marker)
-		logPath := ServerRunLogPath(runID, serverID)
-		_ = StepLogs.Open(commandID_placeholder, logPath, secretsSlice(secretVals))
-```
-
-There is a chicken-and-egg with `commandID`: today `dispatchAndWait` generates the `commandID` internally. Refactor so the runner owns the `commandID`:
-
-1. Change `dispatchAndWait(serverID string, cmd *pb.RunStepCmd)` to `dispatchAndWait(serverID, commandID string, cmd *pb.RunStepCmd)` and remove its internal `commandID := uuid.New().String()` (use the passed one).
-2. In `runServer`, generate `commandID := uuid.New().String()` at the top of each attempt-group (before the marker/open), open the log with it, then call `dispatchAndWait(serverID, commandID, cmd)`.
-3. After the step completes (result received), call `StepLogs.Close(commandID)` defensively (idempotent — the agent's eof usually closed it already; Close on a missing key is a no-op).
-
-Add a helper to convert the `secretVals map[string]string` to a `[]string` of values:
-
-```go
-func secretsSlice(m map[string]string) []string {
-	out := make([]string, 0, len(m))
-	for _, v := range m {
-		out = append(out, v)
-	}
-	return out
-}
-```
-
-Update `finishStep(...)` call + signature: **remove** the `stdout, stderr string` params and the `output_env` masking stays. Persist `log_offset` instead. New `finishStep`:
-
-```go
-func finishStep(runID, serverID string, order int, status string, attempts, exit int, logOffset int64, outEnv map[string]string) {
-	now := time.Now()
-	updateStep(runID, serverID, order, bson.M{
-		"server_runs.$[s].steps.$[t].status":      status,
-		"server_runs.$[s].steps.$[t].attempts":    attempts,
-		"server_runs.$[s].steps.$[t].exit_code":   exit,
-		"server_runs.$[s].steps.$[t].log_offset":  logOffset,
-		"server_runs.$[s].steps.$[t].output_env":  outEnv,
-		"server_runs.$[s].steps.$[t].finished_at": now,
-	})
-}
-```
-
-In the loop, after receiving `res`, drop the `stdout, stderr := ...` masking of `res.Stdout/res.Stderr` (those are now streamed to file). Keep the `outEnv` build **with existing masking** (`maskSecrets(v, allSecrets)` per the merged secret-leak fix) — `output_env`/`run_env` masking is unchanged. Call:
-
-```go
-		finishStep(runID, serverID, i, status, attempts, exit, offset, outEnv)
-```
-
-where `offset` is the marker offset captured before dispatch. If `res == nil`, still write a short note to the file so failures are visible:
-
-```go
-		if res == nil {
-			_, _ = AppendMarker(runID, serverID, "[vantage] agent did not return a result\n")
-		}
-```
-
-Remove the initial `StepRun{... Status:"queued"}` `Stdout/Stderr` references if any (the model no longer has them — the queued StepRun in `TriggerWorkflow` set only `Order/Name/Status/OutputEnv`, so no change needed there; verify).
-
-Ensure `fmt` is imported (it already is).
-
-- [ ] **Step 4: Verify build**
-
-Run: `cd server && go build ./... && go vet ./...`
-Expected: success. Fix any remaining references to the removed `Stdout`/`Stderr` fields or the old `finishStep`/`dispatchAndWait` signatures.
-
-- [ ] **Step 5: Commit**
-
-```bash
-git add server/internal/models/workflow.go server/internal/services/workflow_runner.go server/internal/grpc/server.go
-git commit -m "feat(server): stream step logs to files, drop log bodies from run docs"
-```
-
----
-
-## Task 6: REST — log fetch + SSE stream endpoints
-
-**Files:**
-- Modify: `server/internal/api/workflows.go`
-
-**Interfaces:**
-- Consumes: `ServerRunLogPath`, `GetRun` (existing).
-- Produces: `GET /api/runs/:runId/servers/:serverId/logs` and `GET /api/runs/:runId/servers/:serverId/logs/stream` (SSE).
-
-- [ ] **Step 1: Add the two handlers + routes**
-
-In `registerWorkflowRoutes`, add:
-
-```go
-	g.GET("/runs/:runId/servers/:serverId/logs", getServerRunLog)
-	g.GET("/runs/:runId/servers/:serverId/logs/stream", streamServerRunLog)
-```
-
-Add a UUID-ish validator and the handlers:
-
-```go
-var uuidLike = regexp.MustCompile(`^[a-zA-Z0-9-]{1,64}$`)
-
-func getServerRunLog(c *gin.Context) {
-	runID, serverID := c.Param("runId"), c.Param("serverId")
-	if !uuidLike.MatchString(runID) || !uuidLike.MatchString(serverID) {
-		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid id"})
-		return
-	}
-	path := services.ServerRunLogPath(runID, serverID)
-	b, err := os.ReadFile(path)
-	if err != nil {
-		c.JSON(http.StatusNotFound, gin.H{"error": "no logs"})
-		return
-	}
-	c.Data(http.StatusOK, "text/plain; charset=utf-8", b)
-}
-
-func streamServerRunLog(c *gin.Context) {
-	runID, serverID := c.Param("runId"), c.Param("serverId")
-	if !uuidLike.MatchString(runID) || !uuidLike.MatchString(serverID) {
-		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid id"})
-		return
-	}
-	path := services.ServerRunLogPath(runID, serverID)
-
-	c.Writer.Header().Set("Content-Type", "text/event-stream")
-	c.Writer.Header().Set("Cache-Control", "no-cache")
-	c.Writer.Header().Set("Connection", "keep-alive")
-	c.Writer.Header().Set("X-Accel-Buffering", "no")
-
-	flusher, ok := c.Writer.(http.Flusher)
-	if !ok {
-		c.JSON(http.StatusInternalServerError, gin.H{"error": "stream unsupported"})
-		return
-	}
-
-	var offset int64
-	sendNew := func() bool {
-		f, err := os.Open(path)
-		if err != nil {
-			return true // file may not exist yet; keep waiting
-		}
-		defer f.Close()
-		if _, err := f.Seek(offset, 0); err != nil {
-			return true
-		}
-		buf := make([]byte, 32*1024)
-		for {
-			n, _ := f.Read(buf)
-			if n <= 0 {
-				break
-			}
-			offset += int64(n)
-			// SSE data frame; split on newlines to keep frames well-formed.
-			for _, line := range splitSSE(buf[:n]) {
-				_, _ = c.Writer.WriteString("data: " + line + "\n")
-			}
-			_, _ = c.Writer.WriteString("\n")
-			flusher.Flush()
-		}
-		return true
-	}
-
-	ctx := c.Request.Context()
-	ticker := time.NewTicker(500 * time.Millisecond)
-	defer ticker.Stop()
-	for {
-		sendNew()
-		if serverRunTerminal(runID, serverID) {
-			sendNew() // final drain
-			_, _ = c.Writer.WriteString("event: done\ndata: end\n\n")
-			flusher.Flush()
-			return
-		}
-		select {
-		case <-ctx.Done():
-			return
-		case <-ticker.C:
-		}
-	}
-}
-
-// serverRunTerminal reports whether the given server-run has reached a terminal status.
-func serverRunTerminal(runID, serverID string) bool {
-	r, err := services.GetRun(runID)
-	if err != nil {
-		return true
-	}
-	for _, sr := range r.ServerRuns {
-		if sr.ServerID == serverID {
-			switch sr.Status {
-			case "success", "failed", "skipped", "cancelled":
-				return true
-			}
-			return false
-		}
-	}
-	return true
-}
-
-// splitSSE turns a raw byte slice into SSE-safe payload lines (newlines become
-// separate data lines; carriage returns stripped).
-func splitSSE(b []byte) []string {
-	s := strings.ReplaceAll(string(b), "\r", "")
-	return strings.Split(s, "\n")
-}
-```
-
-Add imports: `"os"`, `"regexp"`, `"strings"`, `"time"`, `"net/http"` (already present). Confirm `services.GetRun` and `ServerRun.Status`/`ServerID` fields exist (they do from the Workflows feature).
-
-- [ ] **Step 2: Verify build**
-
-Run: `cd server && go build ./... && go vet ./...`
-Expected: success.
-
-- [ ] **Step 3: Commit**
-
-```bash
-git add server/internal/api/workflows.go
-git commit -m "feat(api): server-run log fetch and SSE stream endpoints"
-```
-
----
-
-## Task 7: Frontend — live SSE tail + retention setting
-
-**Files:**
-- Modify: `web/lib/api.ts`
-- Modify: `web/app/workflows/[id]/runs/[runId]/page.tsx`
-- Modify: `web/app/settings/page.tsx`
-
-**Interfaces:**
-- Consumes: SSE endpoint, logs endpoint, settings mutation.
-
-- [ ] **Step 1: Update API types + helpers**
-
-In `web/lib/api.ts`:
-- In `StepRun`, remove `stdout` and `stderr`; add `log_offset: number`.
-- Add: `getServerRunLog: (runId: string, serverId: string) => request<string>(...)` — but the logs endpoint returns `text/plain`, so add a dedicated fetch that reads text. If `request<T>` assumes JSON, add a sibling:
-
-```ts
-  async getServerRunLog(runId: string, serverId: string): Promise<string> {
-    const res = await fetch(`${API_BASE}/api/runs/${runId}/servers/${serverId}/logs`, { credentials: "include" });
-    if (!res.ok) throw new Error("no logs");
-    return res.text();
-  },
-```
-
-(Use the file's real base-URL constant / credentials pattern — inspect how `request` builds URLs and mirror it. If the app is same-origin with a rewrite, a relative `/api/...` fetch is fine.)
-- Export a helper to build the SSE URL: `serverRunLogStreamUrl(runId, serverId)` returning the `/api/runs/:runId/servers/:serverId/logs/stream` URL against the same base.
-- In the Settings type, add `workflow_log_retention_days?: number | null`.
-
-- [ ] **Step 2: Live tail in the run detail page**
-
-In `web/app/workflows/[id]/runs/[runId]/page.tsx`:
-- Remove all use of `st.stdout` / `st.stderr` (fields gone). Step `<details>` now show status/exit/attempts pills only.
-- Add a per-server live terminal. For each `server_run`, render a `<pre>` and, while `sr.status === "running"`, subscribe via `EventSource`:
-
-```tsx
-function ServerLog({ runId, serverId, status }: { runId: string; serverId: string; status: string }) {
-  const [text, setText] = useState("");
-  const preRef = useRef<HTMLPreElement>(null);
-  const running = status === "running";
-
-  useEffect(() => {
-    if (running) {
-      const es = new EventSource(api.serverRunLogStreamUrl(runId, serverId), { withCredentials: true });
-      es.onmessage = (e) => setText((t) => t + e.data + "\n");
-      es.addEventListener("done", () => es.close());
-      es.onerror = () => es.close();
-      return () => es.close();
-    }
-    // terminal: fetch the whole file once
-    api.getServerRunLog(runId, serverId).then(setText).catch(() => setText(""));
-  }, [running, runId, serverId]);
-
-  useEffect(() => { preRef.current?.scrollTo(0, preRef.current.scrollHeight); }, [text]);
-
-  return (
-    <pre ref={preRef} className="mt-2 max-h-80 overflow-auto rounded bg-black/40 p-2 font-mono text-xs text-text-secondary whitespace-pre-wrap">
-      {text || (running ? "Waiting for output…" : "No output.")}
-    </pre>
-  );
-}
-```
-
-Render `<ServerLog runId={run.run_id} serverId={sr.server_id} status={sr.status} />` inside each server card, below the step pills. Keep the existing react-query `refetchInterval` on the run (drives status pills); the SSE handles live text.
-
-- [ ] **Step 3: Retention field in Settings**
-
-In `web/app/settings/page.tsx`, add a "Workflow log retention (days)" number input bound to `workflow_log_retention_days`, saved through the existing settings save mutation. Add helper text: "0 = keep forever." Match the page's existing input styling.
-
-- [ ] **Step 4: Verify build**
-
-Run: `cd web && npm run build`
-Expected: type-checks and builds. Fix any lingering `st.stdout`/`st.stderr` references.
-
-- [ ] **Step 5: Commit**
-
-```bash
-git add web/lib/api.ts web/app/workflows/[id]/runs/[runId]/page.tsx web/app/settings/page.tsx
-git commit -m "feat(web): live SSE log tail and log retention setting"
-```
-
----
-
-## Task 8: End-to-end verification
-
-**Files:** none (verification only).
-
-- [ ] **Step 1: Build everything**
-
-Run: `cd server && go build ./... && go vet ./... && cd ../agent && go build ./... && go vet ./... && cd ../web && npm run build`
-Expected: all succeed.
-
-- [ ] **Step 2: Manual smoke (documented, run if an environment is available)**
-
-With server + MongoDB + a connected agent:
-1. Run a workflow with a step that emits output slowly (e.g. `for i in $(seq 1 10); do echo "line $i"; sleep 1; done`). Open the run detail page while running; confirm lines appear live (SSE), not only at the end.
-2. Confirm `<logdir>/<run_id>/<server_id>.log` exists on the server with step markers and the output.
-3. Confirm `workflow_runs` doc no longer stores stdout/stderr bodies; `steps[].log_offset` is set.
-4. Add a secret ref and echo it; confirm the file shows `***`, including when the secret would straddle a chunk boundary.
-5. Set retention to 0 in Settings → confirm sweeper keeps files; set to a small value and backdate a run's `finished_at` → confirm the dir is removed within the hour (or call `sweepLogs` path manually).
-
-- [ ] **Step 3: Commit any fixes found**
-
-```bash
-git add -A
-git commit -m "fix: workflow log streaming e2e fixes"
-```
-
----
-
-## Self-Review Notes
-
-- **Spec coverage:** §3 proto → T1; §4 agent streaming → T2; §5.1 registry + §7 sweeper → T3; §7.1 setting + startup → T4; §5.3/§5.4 runner+model → T5; §6 REST/SSE → T6; §8 frontend → T7. Tests omitted per Global Constraints.
-- **Masking** boundary-safe carry buffer in `StepLogs.Append`, flushed in `Close` (T3); `output_env`/`run_env` masking unchanged (T5 keeps the merged fix).
-- **commandID ownership** moved to the runner so the log file can be opened before dispatch (T5) — mirrors the `StepResults.Await`-before-dispatch ordering.
-- **Buildable commits:** T3 adds a temp stub for `GetWorkflowLogRetentionDays`, removed in T4.
-- **Removed fields** `StepRun.Stdout/Stderr` — every reader updated in T5 (runner) and T7 (frontend).
-- **Open follow-ups (out of scope):** per-step SSE channels, log download/zip, compression, pre-existing runs have no files.
-```

docs/superpowers/specs/2026-07-17-web-console-design.md

@@ -1,241 +0,0 @@
-# Vantage Web Console (Guacamole Replacement) — Design
-
-**Date:** 2026-07-17
-**Status:** Approved design, pre-implementation
-
-## Goal
-
-Add a browser-based remote-access console to Vantage — SSH, RDP, and VNC into
-managed servers — as a self-hosted Guacamole replacement. Users select an SSH
-key to connect over SSH. RDP targets are reachable from a new Windows agent that
-registers the host and reports status. Windows agent ships as an MSI installer
-produced by CI.
-
-## Non-Goals (YAGNI)
-
-- Session recording / replay (may be added later).
-- Native Go RDP implementation (guacd handles protocol translation).
-- Per-user Linux/Windows account management from the agent.
-- Tunneling console traffic through the agent (direct network path assumed).
-
----
-
-## Architecture
-
-```
-Browser (guacamole-common-js, vendored — no CDN)
-   │  Guacamole protocol over WebSocket
-   ▼
-Go server: /api/console/tunnel   (github.com/wwt/guac)
-   │  Guacamole protocol over TCP :4822
-   ▼
-guacd container (Apache Guacamole daemon)
-   │  SSH :22 / RDP :3389 / VNC :5900  — direct to target IP
-   ▼
-Target host (LAN / VPN line-of-sight from server)
-```
-
-- **Browser:** loads vendored `guacamole-common-js`, renders RDP/VNC display and
-  SSH terminal. No external CDN (matches existing infra rules).
-- **Go server:** exposes a WebSocket tunnel endpoint using `github.com/wwt/guac`
-  (Go Guacamole tunnel library). No Java `guacamole-client` required.
-- **guacd:** new container in `deploy/docker-compose.yml`, bound to the internal
-  docker network only, reachable by the server on `:4822`.
-- **Network path:** guacd connects **directly** to the target IP. Requires the
-  central server to have network line-of-sight to hosts (homelab LAN / VPN). The
-  agent's outbound-only guarantee is unchanged — the console path is
-  server→target, not agent-mediated.
-
----
-
-## Data Model Changes
-
-### `keys` — extend to hold private material
-
-```json
-{
-  "key_id": "uuid",
-  "label": "dom-macbook",
-  "public_key": "ssh-ed25519 AAAA...",
-  "private_key_enc": "<AES-256-GCM ciphertext | null>",
-  "has_private": true,
-  "passphrase_enc": "<AES-256-GCM ciphertext | null>",
-  "fingerprint": "SHA256:...",
-  "source": "uploaded|generated",
-  "created_at": "ISODate"
-}
-```
-
-- A key may be created from an uploaded **private+public** pair, upload of a
-  public key only, or agent generation.
-- Agent key generation now also uploads `private_key_enc` (reuses the existing
-  AES-256 key used for at-rest encryption). Private key no longer stays local
-  only — it is stored encrypted so the console can reuse it.
-- Optional `passphrase_enc` for passphrase-protected private keys.
-- Console lists only keys where `has_private = true`.
-
-### `servers` — extend with console metadata
-
-```json
-{
-  "...": "...existing fields...",
-  "os_type": "linux|windows",
-  "console_protocols": ["ssh"],
-  "ssh_port": 22,
-  "rdp_port": 3389
-}
-```
-
-- `os_type` set at registration from the agent.
-- `console_protocols` lists enabled protocols per server (`ssh`, `rdp`, `vnc`).
-- Port fields default to standard ports, overridable in the UI.
-
-### `console_sessions` — new collection (audit)
-
-```json
-{
-  "session_id": "uuid",
-  "server_id": "uuid",
-  "protocol": "ssh|rdp|vnc",
-  "key_id": "uuid | null",
-  "user": "who opened it",
-  "started_at": "ISODate",
-  "ended_at": "ISODate | null",
-  "client_ip": "string"
-}
-```
-
----
-
-## Session Broker + Connection Flow
-
-New service: `server/internal/services/console.go`.
-
-1. Browser `POST /api/console/connect`
-   `{ server_id, protocol, key_id?, rdp_username?, rdp_password? }`.
-2. Broker validates request, loads the server (host IP, port for protocol),
-   loads the key and **decrypts `private_key_enc` in memory only**.
-3. Builds the guacd connection parameter map:
-   - **SSH:** `hostname`, `port`, `username`, `private-key` (decrypted),
-     `passphrase` (if any).
-   - **RDP:** `hostname`, `port`, `username`, `password`, `security=any`,
-     `ignore-cert=true`.
-   - **VNC:** `hostname`, `port`, `password`.
-4. Creates a `console_sessions` document, returns a short-lived signed session
-   token.
-5. Browser opens WebSocket `/api/console/tunnel?token=…`. The `wwt/guac` handler
-   validates the token, dials guacd `:4822`, and pipes bytes in both directions.
-6. On socket close, the broker sets `ended_at` on the session doc.
-
-### Security
-
-- Decrypted private keys and RDP passwords are **never persisted, never logged,
-  never sent to the browser** — passed only to guacd.
-- Session token: short TTL (~60s to open the WebSocket), single-use,
-  HMAC-signed, bound to the authenticated user.
-- guacd is bound to the internal docker network only; not exposed publicly.
-- At-rest encryption (`private_key_enc`, `passphrase_enc`) reuses the existing
-  AES-256 key already used for agent-generated private keys.
-
----
-
-## Windows Agent
-
-Same Go codebase as the Linux agent, with a reduced role: **register +
-heartbeat + status only**. No `authorized_keys` management (meaningless on
-Windows).
-
-- Build target: `GOOS=windows GOARCH=amd64` → `vantage-agent-windows-amd64.exe`.
-- Agent detects OS at registration and sends `os_type=windows`.
-- The key-sync loop is disabled on Windows via a runtime OS check (or build tag)
-  — no `authorized_keys` writes are ever attempted.
-- Config file: `C:\ProgramData\vantage\config.yaml`, locked down via ACL to the
-  equivalent of `0600`.
-- Runs as a Windows service via **nssm**.
-
----
-
-## Windows Installer (MSI)
-
-Agent ships as a WiX v4 MSI produced in CI.
-
-- **WiX v4** chosen because it is a dotnet tool that builds MSIs
-  **cross-platform** — runs on the Linux Gitea act_runner. (Inno Setup is
-  Windows-only and does not fit the runner.)
-- MSI bundles `vantage-agent.exe`, installs it to `C:\Program Files\Vantage\`,
-  and registers the nssm service (ships nssm or uses a CustomAction).
-- Accepts install parameters as MSI properties for silent/headless install:
-  ```
-  msiexec /i vantage-agent.msi /qn SERVERID=<id> TOKEN=<token> SERVERURL=vantage..:9090
-  ```
-- GUI install (double-click) prompts for server-id / token / server-url via a
-  dialog.
-
-### Two install paths
-
-1. **Installer direct** — user downloads `vantage-agent.msi`, double-clicks,
-   fills the dialog. No script required.
-2. **PowerShell one-liner** — served dynamically (like the existing bash
-   `/install`). Script downloads the `.msi`, verifies SHA-256, then runs
-   `msiexec /qn` with injected `SERVERID` / `TOKEN` / `SERVERURL`. Used by the
-   copy-paste "Add Server" flow.
-
-The PowerShell script (`/install.ps1`) steps:
-1. Detect arch.
-2. Download `vantage-agent.msi` from the latest Gitea `agent/v*` release.
-3. Verify SHA-256 against `checksums.txt`.
-4. Run `msiexec /i vantage-agent.msi /qn SERVERID=.. TOKEN=.. SERVERURL=..`.
-
----
-
-## Frontend Routes
-
-| Route                     | Change                                                        |
-| ------------------------- | ------------------------------------------------------------- |
-| `/servers`                | Show `os_type` badge, enabled console protocols               |
-| `/servers/[id]`           | Add **Connect** button(s) per enabled protocol                |
-| `/servers/[id]/console`   | New — full-screen console (guacamole-common-js), key picker   |
-| `/servers/new`            | Offer Windows (MSI) vs Linux (bash) install instructions      |
-
-Console page: select protocol + SSH key (SSH) or enter RDP credentials, call
-`/api/console/connect`, open the tunnel WebSocket, mount the Guacamole client.
-
----
-
-## CI/CD Changes
-
-### `agent-release.yml`
-
-- Add `windows/amd64` build: `vantage-agent-windows-amd64.exe`.
-- Add WiX v4 MSI build job → `vantage-agent.msi`.
-- Add both to `checksums.txt` and release assets.
-
-Release assets become:
-- `vantage-agent-linux-amd64`
-- `vantage-agent-linux-arm64`
-- `vantage-agent-windows-amd64.exe`
-- `vantage-agent.msi`
-- `checksums.txt`
-
-### `server-deploy.yml`
-
-- Add guacd service to `deploy/docker-compose.yml` (deployed alongside server).
-
----
-
-## New Dependencies
-
-- **Go:** `github.com/wwt/guac` (Guacamole tunnel/WebSocket in Go).
-- **Container:** `guacamole/guacd` official image.
-- **Frontend:** vendored `guacamole-common-js` (no CDN).
-- **CI:** WiX v4 dotnet tool; nssm binary bundled for the MSI.
-
----
-
-## Open Implementation Notes
-
-- Confirm `wwt/guac` API surface for connection-parameter passing and token auth
-  binding during implementation.
-- nssm packaging inside MSI: bundle the nssm binary as a payload + CustomAction,
-  or run `sc.exe`-based service install if nssm proves awkward in WiX.
-- ACL hardening of `C:\ProgramData\vantage\config.yaml` in the MSI CustomAction.

docs/superpowers/specs/2026-07-20-server-workflows-design.md

@@ -1,238 +0,0 @@
-# Server Workflows — Design
-
-**Date:** 2026-07-20
-**Status:** Approved (design) — ready for implementation planning
-**Scope:** Server Workflows only. Fleet Inventory and SaaS/local-auth are separate sub-projects with their own specs.
-
-Approved UI mockup: three-pane builder (Step Library · Canvas · Inspector), env vars shown riding the wire between nodes.
-
----
-
-## 1. Summary
-
-Let operators compose **reusable shell steps** (Bash or PowerShell) into **workflows** and run them across many managed servers in parallel. Steps pass data to later steps through a `$WORKFLOW_ENV` file (GitHub-Actions style). Every run is recorded with full per-step logs. Steps can reference org secrets, injected as environment variables at runtime.
-
-Builds directly on the existing `CommandStream` gRPC infrastructure (`dispatch.go`, `ServerCommand` oneof, agent command loop).
-
----
-
-## 2. Locked decisions
-
-| Topic | Decision |
-|-------|----------|
-| Data passing | Implicit. Every step's `$WORKFLOW_ENV` outputs merge into the run's env and are exposed to **all** later steps as `$KEY`. No explicit port wiring. |
-| Failure model | Per-step policy: `stop` (default), `continue`, `retry` (with max attempt count). |
-| Targets | Fan-out. Same step sequence runs on N target servers **in parallel**. Steps within one server run **sequentially**. |
-| History/logs | Every run persisted: status, timing, per-server per-step stdout/stderr/exit code, captured output env. |
-| Secrets | Steps declare needed secret keys; resolved from existing `secrets` store and injected as env vars at exec time. Never persisted into run logs. |
-| Testing | **Skipped** for this iteration per request. No test files written. |
-
----
-
-## 3. Data model (MongoDB)
-
-### `workflow_steps` — reusable step library
-```json
-{
-  "_id": "ObjectId",
-  "step_id": "uuid",
-  "name": "Restart service",
-  "description": "Restart-Service by name, wait ready",
-  "interpreter": "bash | powershell",
-  "script": "Restart-Service vantage-api\n...",
-  "declared_outputs": ["STARTED_AT"],     // documentation/UI hints; not enforced
-  "secret_refs": ["DEPLOY_TOKEN"],         // secret keys this step needs injected
-  "org_id": "uuid",                        // for future multi-tenant; single-org for now
-  "created_at": "ISODate",
-  "updated_at": "ISODate"
-}
-```
-
-### `workflows` — ordered composition
-```json
-{
-  "_id": "ObjectId",
-  "workflow_id": "uuid",
-  "name": "Deploy & Restart API",
-  "target_server_ids": ["uuid", "uuid"],
-  "steps": [
-    {
-      "step_id": "uuid",          // reference to library step
-      "order": 0,
-      "on_failure": "stop | continue | retry",
-      "max_retries": 0,           // used when on_failure = retry
-      "overrides": {              // optional local fork of the library step
-        "script": null,
-        "secret_refs": null
-      }
-    }
-  ],
-  "created_at": "ISODate",
-  "updated_at": "ISODate"
-}
-```
-Editing a library step from the Inspector writes an `overrides` block on that workflow step (a local fork) rather than mutating the shared step.
-
-### `workflow_runs` — execution records
-```json
-{
-  "_id": "ObjectId",
-  "run_id": "uuid",
-  "workflow_id": "uuid",
-  "workflow_snapshot": { },        // frozen copy of workflow + resolved steps at trigger time
-  "status": "running | success | failed | cancelled",
-  "triggered_by": "user-id",
-  "started_at": "ISODate",
-  "finished_at": "ISODate | null",
-  "server_runs": [
-    {
-      "server_id": "uuid",
-      "status": "queued | running | success | failed | skipped",
-      "started_at": "ISODate | null",
-      "finished_at": "ISODate | null",
-      "run_env": { "VERSION": "a1b9f0" },   // accumulated non-secret output env
-      "steps": [
-        {
-          "order": 0,
-          "name": "Git pull & build",
-          "status": "success | failed | running | queued | skipped",
-          "attempts": 1,
-          "exit_code": 0,
-          "stdout": "…",
-          "stderr": "…",
-          "output_env": { "VERSION": "a1b9f0" },
-          "started_at": "ISODate",
-          "finished_at": "ISODate"
-        }
-      ]
-    }
-  ]
-}
-```
-Secret values are never written to `stdout`/`stderr`/`run_env` by us; masking of known secret values in captured output is applied before persistence.
-
----
-
-## 4. gRPC protocol changes (`proto/vantage/v1/vantage.proto`)
-
-### New command in the `ServerCommand` oneof
-```protobuf
-message RunStepCmd {
-  string interpreter          = 1; // "bash" | "powershell"
-  string script               = 2;
-  map<string, string> env     = 3; // inputs = accumulated run env + injected secrets
-  int32  timeout_seconds      = 4;
-}
-```
-Add `RunStepCmd run_step = 6;` to the `ServerCommand` oneof.
-
-### Richer result — new `AgentMessage` payload
-Current `CommandResult{command_id, success, message}` is too thin. Add a dedicated step result:
-```protobuf
-message StepResult {
-  string command_id        = 1;
-  int32  exit_code         = 2;
-  string stdout            = 3;
-  string stderr            = 4;
-  map<string, string> output_env = 5; // parsed $WORKFLOW_ENV KEY=value lines
-}
-```
-Add `StepResult step_result = 5;` to the `AgentMessage` oneof (alongside existing `ready` / `result`).
-
----
-
-## 5. Agent execution (`agent/internal/...`)
-
-New handler for `RunStepCmd` in the agent command loop:
-
-1. Create a temp dir; create empty `WORKFLOW_ENV` file inside it.
-2. Write `script` to a temp script file.
-3. Build the process environment: inherited env + `cmd.env` (run env + secrets) + `WORKFLOW_ENV=<path to env file>`.
-4. Execute:
-   - `bash`  → `bash <script>`
-   - `powershell` → `pwsh -NoProfile -File <script>` (fallback `powershell.exe` on Windows if `pwsh` absent).
-5. Capture stdout, stderr, exit code. Enforce `timeout_seconds` (kill on exceed → non-zero exit, stderr note).
-6. Parse the `WORKFLOW_ENV` file: each `KEY=value` line becomes an `output_env` entry (last write wins; supports multi-line via simple `KEY<<EOF` heredoc form, optional for v1 — start with single-line `KEY=value`).
-7. Reply with `StepResult`. Delete temp dir.
-
-Agent runs as root (existing), so no privilege change. Script content is trusted operator input.
-
----
-
-## 6. Server orchestration (`server/internal/services/workflows.go`)
-
-Runner responsibilities:
-
-1. On trigger: snapshot the workflow (resolve each library step + overrides), create a `workflow_runs` doc with one `server_run` per target, all `queued`.
-2. Spawn one goroutine **per target server** (parallel fan-out). Each goroutine:
-   - Verifies the agent is connected (`Dispatcher.IsConnected`); if not → `server_run.status = skipped`, reason recorded.
-   - Maintains a `run_env map[string]string`, seeded empty.
-   - For each step in order:
-     - Resolve `secret_refs` from the secrets service → merge into the command env (kept separate from persisted `run_env`).
-     - Dispatch `RunStepCmd{env: run_env + secrets}` via a **correlated** send — needs a way to await the matching `StepResult` by `command_id` (see §7).
-     - On result: persist step record (stdout/stderr/exit, masked); merge `output_env` into `run_env`.
-     - Apply `on_failure` on non-zero exit: `stop` (fail server_run, break), `continue` (mark failed, proceed), `retry` (re-dispatch up to `max_retries`).
-3. Aggregate: run `status = success` if all server_runs succeeded, else `failed`. Set `finished_at`.
-
-### Concurrency / queue
-- One workflow run per workflow at a time (reject or queue concurrent triggers — v1: reject with clear error).
-- Per-server step dispatch is serial; servers are parallel.
-
----
-
-## 7. Correlated command results
-
-The existing dispatcher is fire-and-forget; workflows need request/response by `command_id`. Add a small **pending-result registry** alongside `Dispatcher`:
-
-- `AwaitResult(commandID) <-chan *pb.StepResult` — registers a channel before dispatch.
-- The `CommandStream` receive loop, on a `StepResult`, looks up the pending channel by `command_id` and delivers it (falls back to existing `CommandResult` handling for other command types).
-- Timeout guard on the server side (step `timeout_seconds` + grace) so a dead agent can't hang a run.
-
-This is additive; existing `CommandResult` flow for key/update commands is unchanged.
-
----
-
-## 8. REST API (`server/internal/api/workflows.go`)
-
-| Method + path | Purpose |
-|---------------|---------|
-| `GET /api/steps` / `POST` / `PUT /:id` / `DELETE /:id` | Reusable step library CRUD |
-| `GET /api/workflows` / `POST` / `PUT /:id` / `DELETE /:id` | Workflow CRUD (name, targets, ordered steps) |
-| `POST /api/workflows/:id/run` | Trigger a run; returns `run_id` |
-| `GET /api/workflows/:id/runs` | Run history (summary list) |
-| `GET /api/runs/:run_id` | Full run detail incl. per-server per-step logs |
-| `POST /api/runs/:run_id/cancel` | Best-effort cancel |
-
-Secrets are referenced by key only through these APIs; values never returned.
-
----
-
-## 9. Frontend (`web/app/workflows/`)
-
-- `/workflows` — list workflows, last run status/time, Run button.
-- `/workflows/[id]` — the three-pane builder from the approved mockup:
-  - **Library** (left): reusable steps, `bash`/`pwsh` badges, search, add.
-  - **Canvas** (center): ordered nodes, env chips on wires, live status pills.
-  - **Inspector** (right): name, command editor, declared inputs/outputs, `secret_refs` picker, `on_failure` + retry count.
-- `/workflows/[id]/runs/[runId]` — run detail: per-server columns, expandable per-step stdout/stderr, exit codes, timing. Live-updating while `running` (poll, consistent with existing 30s-poll ethos — or reuse whatever the console screen uses).
-
-Reuse existing web components/styling patterns (there is already `servers`, `secrets`, `audit`, console UI to match).
-
----
-
-## 10. Security notes
-
-- Scripts are trusted operator input executed as root — same trust level as the existing console feature. No new sandbox in v1.
-- Secret values injected as env only; masked from all persisted logs (`stdout`/`stderr`/`run_env`) by literal replacement before write.
-- Run triggering and step/workflow CRUD gated behind existing auth (`server/internal/auth`).
-- Audit: emit audit-log entries (existing `audit` service) on workflow create/edit/delete and run trigger.
-
----
-
-## 11. Out of scope (this iteration)
-
-- Tests (explicitly skipped).
-- Branching/conditional steps, matrix per-server conditionals (fan-out only).
-- Scheduled/cron triggers (manual run only for v1).
-- Multi-org isolation enforcement (schema carries `org_id` for later; single-org behavior now).
-- Artifact upload/collection beyond env vars.

docs/superpowers/specs/2026-07-20-workflow-log-streaming-design.md

@@ -1,193 +0,0 @@
-# Workflow Log Streaming — Design
-
-**Date:** 2026-07-20
-**Status:** Approved (design) — ready for implementation planning
-**Scope:** Stream step stdout/stderr live from agent to server-side log files, tail them live in the UI, and auto-expire them on a retention period. Enhancement to the already-merged Server Workflows feature. No auth/orgs, no inventory.
-
----
-
-## 1. Summary
-
-Today a workflow step buffers all stdout/stderr in agent RAM, ships it in one terminal `StepResult`, and the server persists the whole body into the `workflow_runs` Mongo document. Long/chatty steps risk: agent memory blow-up, the gRPC 4MB message ceiling, and the Mongo 16MB document cap.
-
-Change to **live streaming**:
-
-1. Agent streams output chunks over the existing `CommandStream` as the process runs.
-2. Server appends chunks (secret-masked) to a **per-server-run log file** on disk — not Mongo.
-3. UI tails the file live via **SSE** while a server-run is running; slices per-step by byte offset after completion.
-4. A **retention sweeper** deletes old run-log directories on a configurable period (default 30 days, set in Settings).
-
-`workflow_runs` documents shrink: they no longer carry `stdout`/`stderr` bodies, only status/exit/attempts/output_env/timestamps plus a per-step `log_offset`.
-
----
-
-## 2. Locked decisions
-
-| Topic | Decision |
-|-------|----------|
-| Transport | Reuse bidirectional `CommandStream`. New `AgentMessage.StepOutput` chunk message. |
-| Chunk shape | `{command_id, seq, data, eof}`. Interleaved stdout+stderr in execution order. |
-| Terminal result | `StepResult` still sent at step end, now carries only `exit_code` + `output_env` (no stdout/stderr). |
-| Log granularity | **One file per server-run**: `<logdir>/<run_id>/<server_id>.log`, with a marker line before each step. |
-| Streams | **Interleaved** — single synchronized writer on the agent, terminal-order output. |
-| Masking | **Server-side** (agent can't tell secret env from normal env). Per-stream carry buffer of `maxSecretLen-1` bytes so a secret split across a chunk boundary still masks; flushed on EOF. |
-| Live tail | **SSE** at per-server-run granularity: `GET /api/runs/:runId/servers/:serverId/logs/stream`. Post-run whole-file fetch + per-step offset slice. |
-| Retention | `settings.workflow_log_retention_days`, default **30**, editable in `/settings`. Hourly sweeper deletes `<logdir>/<run_id>/` dirs older than retention by run `finished_at`. |
-| Log dir | Env `VANTAGE_WORKFLOW_LOG_DIR`, default `<data>/workflow-logs`. Created `0700`. |
-| Mongo | No log bodies in `workflow_runs`. Disk is the source of truth for output. |
-
----
-
-## 3. gRPC protocol (`proto/vantage/v1/vantage.proto` + both `pb.go` files)
-
-Add to the `AgentMessage` oneof: `StepOutputChunk step_output = 6;`
-
-```protobuf
-message StepOutputChunk {
-  string command_id = 1;
-  uint64 seq        = 2;   // monotonic per command_id, 0-based
-  bytes  data       = 3;   // raw interleaved stdout+stderr bytes
-  bool   eof        = 4;   // true on the final (empty) chunk
-}
-```
-
-`StepResult` is unchanged in shape but `stdout`/`stderr` are now left empty by the agent (kept in the message for backward-compat / error notes only — server ignores them for log content). The server still reads `exit_code` and `output_env` from `StepResult`.
-
-Hand-written JSON-codec struct added to **both** `server/internal/grpc/pb/vantage.pb.go` and `agent/internal/grpc/pb/vantage.pb.go`, identical. `AgentMessage` gains `StepOutput *StepOutputChunk` in both.
-
-`data` is `[]byte` in the Go structs (JSON-codec base64-encodes it, which is fine).
-
----
-
-## 4. Agent (`agent/internal/exec/exec.go`)
-
-`RunStep` signature gains a chunk sink:
-
-```go
-func RunStep(cmd *pb.RunStepCmd, emit func(seq uint64, data []byte)) *pb.StepResult
-```
-
-- Replace the two `bytes.Buffer`s with a single `streamWriter` set as **both** `c.Stdout` and `c.Stderr`. Its `Write` takes a mutex (so stdout+stderr interleave without interleaving *within* a write), assigns the next `seq`, and calls `emit(seq, copyOfBytes)`. Chunks are whatever the OS pipe delivers (typically ≤64KB); no extra buffering/line-assembly.
-- `StepResult` returns with `Stdout`/`Stderr` empty; `ExitCode` and `OutputEnv` populated as today (env parsing unchanged).
-- On timeout/exec error, put the short note in `StepResult.Stderr` (terminal, not streamed) so the runner can still surface a failure reason even if nothing streamed.
-
-Agent loop (`agent/internal/sync/sync.go`, the `cmd.RunStep != nil` goroutine): pass an `emit` closure that sends `AgentMessage{ServerId, AgentToken, StepOutput: &pb.StepOutputChunk{CommandId, Seq, Data}}` through the existing mutex-guarded `send()`. After `RunStep` returns, send a final `StepOutput{eof:true, seq:last+1}` then the terminal `StepResult` (both via `send()`). Ordering: all chunks, then eof, then StepResult.
-
----
-
-## 5. Server write path
-
-### 5.1 Log writer registry (`server/internal/services/steplogs.go`)
-
-Parallel to `StepResults`. Keyed by `command_id`:
-
-```go
-type stepLogWriter struct {
-  f         *os.File
-  mu        sync.Mutex
-  carry     []byte            // held-back tail for boundary-safe masking
-  secrets   []string          // secret literals to mask
-  maxSecret int
-}
-var StepLogs = &stepLogRegistry{ ... }
-func (r *stepLogRegistry) Open(commandID, path string, secrets []string) (*stepLogWriter, error)
-func (r *stepLogRegistry) Append(commandID string, data []byte)   // masked write
-func (r *stepLogRegistry) Close(commandID string)                 // flush carry, close file
-```
-
-- `Append` masking: concatenate `carry+data`, mask all secret literals (`ReplaceAll(v,"***")`), then write everything except the last `maxSecret-1` bytes; keep those as the new `carry`. `Close` masks+writes the remaining carry. If `secrets` empty, write straight through (no carry).
-- The file handle is opened append-only (`O_APPEND|O_CREATE|O_WRONLY`, `0600`); dir `0700`.
-
-### 5.2 Stream delivery (`server/internal/grpc/server.go`)
-
-In the receive loop, after the `m.StepResult` block, add:
-
-```go
-if m.StepOutput != nil {
-    if m.StepOutput.Eof {
-        services.StepLogs.Close(m.StepOutput.CommandId)
-    } else {
-        services.StepLogs.Append(m.StepOutput.CommandId, m.StepOutput.Data)
-    }
-}
-```
-
-### 5.3 Runner changes (`server/internal/services/workflow_runner.go`)
-
-- Resolve the log dir + run/server file path once per server-run; ensure `<logdir>/<run_id>/` exists.
-- Before dispatching each step: write the step marker line to the file (`\n===== step <order>: <name> =====\n`), record the current file byte offset as the step's `log_offset` (persisted on the `StepRun`), and `StepLogs.Open(commandID, path, secretVals)` **before** `DispatchRunStep` (same ordering rule as `StepResults.Await`).
-- `dispatchAndWait` no longer expects stdout/stderr in the result. On terminal `StepResult`, `StepLogs.Close(commandID)` is driven by the agent's eof; the runner also calls `Close` defensively on timeout/dispatch-failure (idempotent).
-- **Drop** `stdout`/`stderr` from `finishStep` persistence. Masking of the streamed body is done in `Append`; `run_env`/`output_env` masking (existing, from the merged fix) stays.
-- The shared server-run file is written by two writers that never overlap in time (steps are serial, and the runner writes each step marker *before* `StepLogs.Open`): (a) the runner writes markers directly to the path, serially between steps; (b) `StepLogs` writes chunks during a step. **Resolved approach:** `Open(commandID, path, secrets)` opens the path fresh with `O_APPEND|O_CREATE|O_WRONLY` for that step and `Close(commandID)` closes it on eof. One handle live at a time per server-run (serial steps guarantee this), so there is no shared-handle race and no ref-counting. The runner's marker write is a separate short `O_APPEND` open/write/close on the same path.
-
-### 5.4 Data model (`server/internal/models/workflow.go`)
-
-`StepRun`:
-- **Remove** `Stdout`, `Stderr` string fields.
-- **Add** `LogOffset int64 `bson:"log_offset" json:"log_offset"`` — byte offset in the server-run file where this step's marker begins.
-
-`ServerRun` gains nothing structural (its file path is derivable: `<logdir>/<run_id>/<server_id>.log`).
-
----
-
-## 6. REST API (`server/internal/api/workflows.go`)
-
-- `GET /api/runs/:runId/servers/:serverId/logs` — returns the whole server-run log file (`text/plain`). 404 if absent. Used post-run and as SSE fallback.
-- `GET /api/runs/:runId/servers/:serverId/logs/stream` — **SSE**. Opens the file, streams existing content as `data:` events, then polls for appends (~500ms) emitting new bytes, until the server-run status is terminal (success/failed/skipped/cancelled) AND no more bytes, then sends a final `event: done` and closes. Sets `Content-Type: text/event-stream`, disables gin's buffering. Guards against path traversal (runId/serverId are used as literal path segments — validate they are UUIDs / contain no separators).
-
-Log content served by these endpoints is already masked (masking happens at write time), so no masking needed on read.
-
----
-
-## 7. Retention
-
-### 7.1 Setting
-
-`settings` collection gains `workflow_log_retention_days int` (default 30 when unset). Read/write via the existing settings service + surfaced in `/settings` UI as a number input. `0` or negative disables sweeping (keep forever) — document this.
-
-### 7.2 Sweeper (`server/internal/services/steplogs.go` or `logsweeper.go`)
-
-- `StartLogSweeper()` launched at server startup (next to index setup): hourly `time.Ticker`.
-- Each tick: read retention setting; if ≤0 skip. Compute cutoff = `now - retentionDays`. For each `<logdir>/<run_id>/` dir, look up the run's `finished_at` (query `workflow_runs` by run_id); if finished and older than cutoff, `os.RemoveAll` the dir. Fallback to dir mtime if the run doc is gone.
-- Also run once at startup.
-
----
-
-## 8. Frontend
-
-### 8.1 API client (`web/lib/api.ts`)
-
-- `StepRun`: remove `stdout`/`stderr`; add `log_offset: number`.
-- Add `getServerRunLog(runId, serverId): Promise<string>` (GET .../logs).
-- SSE consumed directly via `EventSource` in the component (not through the `request` helper), URL built from the same base.
-- Settings type gains `workflow_log_retention_days`.
-
-### 8.2 Run detail (`web/app/workflows/[id]/runs/[runId]/page.tsx`)
-
-- Per-server card: while the server-run is `running`, open an `EventSource` to the stream endpoint and render a live `<pre>` terminal that appends incoming chunks (auto-scroll). Close the source on `event: done`, unmount, or terminal status.
-- After completion: fetch the whole file once and render it; step `<details>` still list status/exit/attempts pills. (Per-step slicing by `log_offset` is optional polish — v1 may show the whole server log under the card and keep step pills as the status summary.)
-- Remove reliance on `st.stdout`/`st.stderr` (fields gone).
-
-### 8.3 Settings (`web/app/settings/page.tsx`)
-
-- Add a "Workflow log retention (days)" number input bound to `workflow_log_retention_days`, saved via the existing settings mutation. Note that `0` = keep forever.
-
----
-
-## 9. Security
-
-- Secret masking moves to the streaming write path but remains server-side and boundary-safe (carry buffer). Same `***` replacement.
-- Log files `0600`, dirs `0700`, under a dedicated log dir.
-- SSE/read endpoints validate `runId`/`serverId` as UUID-shaped path segments to prevent traversal; they are session-authed (same `apiGroup`).
-- Terminal `StepResult.Stderr` (error notes only) is still masked before any persistence (it is no longer persisted as log body; if surfaced, mask against secretVals).
-
----
-
-## 10. Out of scope
-
-- Per-step (rather than per-server) live SSE channels.
-- Log compression / rotation within a run, remote log storage (S3), download-as-zip.
-- Full-text search over logs.
-- Backfilling/migrating already-existing `workflow_runs` stdout/stderr into files (pre-existing runs keep whatever they had; new field just won't be set — acceptable, feature is new).
-- Tests (skipped, consistent with the Workflows iteration).
-```

proto/vantage/v1/vantage.proto

@@ -92,9 +92,16 @@ message ServerCommand {
     UpdateAgentCmd  update_agent  = 4;
     ApplyUpdatesCmd apply_updates = 5;
     RunStepCmd      run_step      = 6;
+    CleanupWorkspaceCmd cleanup_workspace = 7;
   }
 }
 
+// CleanupWorkspaceCmd tells the agent to recursively remove the run's working
+// directory once all steps on that server have finished.
+message CleanupWorkspaceCmd {
+  string workspace_id = 1;
+}
+
 message DeleteKeyCmd {
   string label = 1;
 }
@@ -117,6 +124,7 @@ message RunStepCmd {
   string script           = 2;
   map<string, string> env = 3;
   int32  timeout_seconds  = 4;
+  string workspace_id     = 5; // per-run working dir the agent creates & uses as cwd
 }
 
 message StepResult {

server/cmd/main.go

@@ -31,6 +31,8 @@ func main() {
 		log.Printf("warning: failed to ensure workflow indexes: %v", err)
 	}
 
+	services.StartLogSweeper()
+
 	redisAddr := getEnv("REDIS_ADDR", "localhost:6379")
 	if err := auth.InitRedis(redisAddr); err != nil {
 		log.Fatalf("failed to connect to Redis: %v", err)

server/internal/api/handlers.go

@@ -450,14 +450,15 @@ func getSettings(c *gin.Context) {
 
 func saveSettings(c *gin.Context) {
 	var body struct {
-		Alerts models.AlertSettings `json:"alerts"`
-		Email  models.EmailSettings `json:"email"`
+		Alerts                   models.AlertSettings `json:"alerts"`
+		Email                    models.EmailSettings `json:"email"`
+		WorkflowLogRetentionDays *int                 `json:"workflow_log_retention_days"`
 	}
 	if err := c.ShouldBindJSON(&body); err != nil {
 		c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
 		return
 	}
-	if err := services.SaveSettings(body.Alerts, body.Email); err != nil {
+	if err := services.SaveSettings(body.Alerts, body.Email, body.WorkflowLogRetentionDays); err != nil {
 		c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
 		return
 	}

server/internal/api/workflows.go

@@ -3,7 +3,11 @@ package api
 import (
 	"fmt"
 	"net/http"
+	"os"
+	"regexp"
 	"strconv"
+	"strings"
+	"time"
 
 	"github.com/gin-gonic/gin"
 	"github.com/mrhid6/vantage/server/internal/models"
@@ -26,6 +30,114 @@ func registerWorkflowRoutes(g *gin.RouterGroup) {
 
 	g.GET("/runs/:runId", getRun)
 	g.POST("/runs/:runId/cancel", cancelRun)
+	g.GET("/runs/:runId/servers/:serverId/logs", getServerRunLog)
+	g.GET("/runs/:runId/servers/:serverId/logs/stream", streamServerRunLog)
+}
+
+var uuidLike = regexp.MustCompile(`^[a-zA-Z0-9-]{1,64}$`)
+
+func getServerRunLog(c *gin.Context) {
+	runID, serverID := c.Param("runId"), c.Param("serverId")
+	if !uuidLike.MatchString(runID) || !uuidLike.MatchString(serverID) {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid id"})
+		return
+	}
+	path := services.ServerRunLogPath(runID, serverID)
+	b, err := os.ReadFile(path)
+	if err != nil {
+		c.JSON(http.StatusNotFound, gin.H{"error": "no logs"})
+		return
+	}
+	c.Data(http.StatusOK, "text/plain; charset=utf-8", b)
+}
+
+func streamServerRunLog(c *gin.Context) {
+	runID, serverID := c.Param("runId"), c.Param("serverId")
+	if !uuidLike.MatchString(runID) || !uuidLike.MatchString(serverID) {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "invalid id"})
+		return
+	}
+	path := services.ServerRunLogPath(runID, serverID)
+
+	c.Writer.Header().Set("Content-Type", "text/event-stream")
+	c.Writer.Header().Set("Cache-Control", "no-cache")
+	c.Writer.Header().Set("Connection", "keep-alive")
+	c.Writer.Header().Set("X-Accel-Buffering", "no")
+
+	flusher, ok := c.Writer.(http.Flusher)
+	if !ok {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": "stream unsupported"})
+		return
+	}
+
+	var offset int64
+	sendNew := func() {
+		f, err := os.Open(path)
+		if err != nil {
+			return // file may not exist yet; keep waiting
+		}
+		defer f.Close()
+		if _, err := f.Seek(offset, 0); err != nil {
+			return
+		}
+		buf := make([]byte, 32*1024)
+		for {
+			n, _ := f.Read(buf)
+			if n <= 0 {
+				break
+			}
+			offset += int64(n)
+			// SSE data frame; split on newlines to keep frames well-formed.
+			for _, line := range splitSSE(buf[:n]) {
+				_, _ = c.Writer.WriteString("data: " + line + "\n")
+			}
+			_, _ = c.Writer.WriteString("\n")
+			flusher.Flush()
+		}
+	}
+
+	ctx := c.Request.Context()
+	ticker := time.NewTicker(500 * time.Millisecond)
+	defer ticker.Stop()
+	for {
+		sendNew()
+		if serverRunTerminal(runID, serverID) {
+			sendNew() // final drain
+			_, _ = c.Writer.WriteString("event: done\ndata: end\n\n")
+			flusher.Flush()
+			return
+		}
+		select {
+		case <-ctx.Done():
+			return
+		case <-ticker.C:
+		}
+	}
+}
+
+// serverRunTerminal reports whether the given server-run has reached a terminal status.
+func serverRunTerminal(runID, serverID string) bool {
+	r, err := services.GetRun(runID)
+	if err != nil {
+		return true
+	}
+	for _, sr := range r.ServerRuns {
+		if sr.ServerID == serverID {
+			switch sr.Status {
+			case "success", "failed", "skipped", "cancelled":
+				return true
+			}
+			return false
+		}
+	}
+	return true
+}
+
+// splitSSE turns a raw byte slice into SSE-safe payload lines (newlines become
+// separate data lines; carriage returns stripped).
+func splitSSE(b []byte) []string {
+	s := strings.ReplaceAll(string(b), "\r", "")
+	return strings.Split(s, "\n")
 }
 
 func listSteps(c *gin.Context) {
@@ -119,7 +231,12 @@ func updateWorkflow(c *gin.Context) {
 		return
 	}
 	services.LogEvent("workflow.updated", actorFromCtx(c), "", c.Param("id"), "workflow updated")
-	c.JSON(http.StatusOK, gin.H{"updated": true})
+	updated, err := services.GetWorkflow(c.Param("id"))
+	if err != nil {
+		c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
+		return
+	}
+	c.JSON(http.StatusOK, updated)
 }
 
 func deleteWorkflow(c *gin.Context) {

server/internal/grpc/pb/vantage.pb.go

@@ -66,12 +66,19 @@ type ReportUpdatesResponse struct{}
 type ApplyUpdatesCmd struct{}
 
 type ServerCommand struct {
-	CommandId    string           `json:"command_id"`
-	GenerateKey  *GenerateKeyCmd  `json:"generate_key,omitempty"`
-	DeleteKey    *DeleteKeyCmd    `json:"delete_key,omitempty"`
-	UpdateAgent  *UpdateAgentCmd  `json:"update_agent,omitempty"`
-	ApplyUpdates *ApplyUpdatesCmd `json:"apply_updates,omitempty"`
-	RunStep      *RunStepCmd      `json:"run_step,omitempty"`
+	CommandId        string               `json:"command_id"`
+	GenerateKey      *GenerateKeyCmd      `json:"generate_key,omitempty"`
+	DeleteKey        *DeleteKeyCmd        `json:"delete_key,omitempty"`
+	UpdateAgent      *UpdateAgentCmd      `json:"update_agent,omitempty"`
+	ApplyUpdates     *ApplyUpdatesCmd     `json:"apply_updates,omitempty"`
+	RunStep          *RunStepCmd          `json:"run_step,omitempty"`
+	CleanupWorkspace *CleanupWorkspaceCmd `json:"cleanup_workspace,omitempty"`
+}
+
+// CleanupWorkspaceCmd tells the agent to recursively remove the run's working
+// directory once all steps on that server have finished.
+type CleanupWorkspaceCmd struct {
+	WorkspaceId string `json:"workspace_id"`
 }
 
 type DeleteKeyCmd struct {
@@ -113,6 +120,9 @@ type RunStepCmd struct {
 	Script         string            `json:"script"`
 	Env            map[string]string `json:"env,omitempty"`
 	TimeoutSeconds int               `json:"timeout_seconds,omitempty"`
+	// WorkspaceId names the per-run working directory the agent creates and uses
+	// as the step's cwd. Empty means run in the agent's default directory.
+	WorkspaceId string `json:"workspace_id,omitempty"`
 }
 
 type StepResult struct {

server/internal/grpc/server.go

@@ -132,6 +132,13 @@ func (s *vantageServer) CommandStream(stream pb.Vantage_CommandStreamServer) err
 			if m.StepResult != nil {
 				services.StepResults.Deliver(m.StepResult)
 			}
+			if m.StepOutput != nil {
+				if m.StepOutput.Eof {
+					services.StepLogs.Close(m.StepOutput.CommandId)
+				} else {
+					services.StepLogs.Append(m.StepOutput.CommandId, m.StepOutput.Data)
+				}
+			}
 		}
 	}()
 

server/internal/models/settings.go

@@ -36,4 +36,6 @@ type Settings struct {
 	Alerts  AlertSettings   `bson:"alerts" json:"alerts"`
 	Email   EmailSettings   `bson:"email" json:"email"`
 	Secrets SecretsSettings `bson:"secrets" json:"secrets"`
+	// WorkflowLogRetentionDays: nil = default 30, 0 = keep forever, N = N days.
+	WorkflowLogRetentionDays *int `bson:"workflow_log_retention_days,omitempty" json:"workflow_log_retention_days,omitempty"`
 }

server/internal/models/workflow.go

@@ -1,26 +1,38 @@
 package models
 
-import "time"
+import (
+	"time"
+
+	"go.mongodb.org/mongo-driver/v2/bson"
+)
+
+type InputParam struct {
+	Name        string `bson:"name" json:"name"`
+	Default     string `bson:"default" json:"default"`
+	Description string `bson:"description" json:"description"`
+}
 
 type WorkflowStep struct {
-	ID              string    `bson:"_id,omitempty" json:"-"`
-	StepID          string    `bson:"step_id" json:"step_id"`
-	Name            string    `bson:"name" json:"name"`
-	Description     string    `bson:"description" json:"description"`
-	Interpreter     string    `bson:"interpreter" json:"interpreter"` // "bash" | "powershell"
-	Script          string    `bson:"script" json:"script"`
-	DeclaredOutputs []string  `bson:"declared_outputs" json:"declared_outputs"`
-	SecretRefs      []string  `bson:"secret_refs" json:"secret_refs"`
-	CreatedAt       time.Time `bson:"created_at" json:"created_at"`
-	UpdatedAt       time.Time `bson:"updated_at" json:"updated_at"`
+	ID              bson.ObjectID `bson:"_id,omitempty" json:"-"`
+	StepID          string        `bson:"step_id" json:"step_id"`
+	Name            string        `bson:"name" json:"name"`
+	Description     string        `bson:"description" json:"description"`
+	Interpreter     string        `bson:"interpreter" json:"interpreter"` // "bash" | "powershell"
+	Script          string        `bson:"script" json:"script"`
+	DeclaredOutputs []string      `bson:"declared_outputs" json:"declared_outputs"`
+	DeclaredInputs  []InputParam  `bson:"declared_inputs" json:"declared_inputs"`
+	SecretRefs      []string      `bson:"secret_refs" json:"secret_refs"`
+	CreatedAt       time.Time     `bson:"created_at" json:"created_at"`
+	UpdatedAt       time.Time     `bson:"updated_at" json:"updated_at"`
 }
 
 type WorkflowStepRef struct {
-	StepID     string        `bson:"step_id" json:"step_id"`
-	Order      int           `bson:"order" json:"order"`
-	OnFailure  string        `bson:"on_failure" json:"on_failure"` // "stop" | "continue" | "retry"
-	MaxRetries int           `bson:"max_retries" json:"max_retries"`
-	Overrides  *StepOverride `bson:"overrides,omitempty" json:"overrides,omitempty"`
+	StepID     string            `bson:"step_id" json:"step_id"`
+	Order      int               `bson:"order" json:"order"`
+	OnFailure  string            `bson:"on_failure" json:"on_failure"` // "stop" | "continue" | "retry"
+	MaxRetries int               `bson:"max_retries" json:"max_retries"`
+	Overrides  *StepOverride     `bson:"overrides,omitempty" json:"overrides,omitempty"`
+	Inputs     map[string]string `bson:"inputs,omitempty" json:"inputs,omitempty"`
 }
 
 type StepOverride struct {
@@ -29,7 +41,7 @@ type StepOverride struct {
 }
 
 type Workflow struct {
-	ID              string            `bson:"_id,omitempty" json:"-"`
+	ID              bson.ObjectID     `bson:"_id,omitempty" json:"-"`
 	WorkflowID      string            `bson:"workflow_id" json:"workflow_id"`
 	Name            string            `bson:"name" json:"name"`
 	TargetServerIDs []string          `bson:"target_server_ids" json:"target_server_ids"`
@@ -44,9 +56,10 @@ type ResolvedStep struct {
 	Name        string   `bson:"name" json:"name"`
 	Interpreter string   `bson:"interpreter" json:"interpreter"`
 	Script      string   `bson:"script" json:"script"`
-	SecretRefs  []string `bson:"secret_refs" json:"secret_refs"`
-	OnFailure   string   `bson:"on_failure" json:"on_failure"`
-	MaxRetries  int      `bson:"max_retries" json:"max_retries"`
+	SecretRefs  []string          `bson:"secret_refs" json:"secret_refs"`
+	OnFailure   string            `bson:"on_failure" json:"on_failure"`
+	MaxRetries  int               `bson:"max_retries" json:"max_retries"`
+	Inputs      map[string]string `bson:"inputs" json:"inputs"`
 }
 
 type StepRun struct {
@@ -55,8 +68,7 @@ type StepRun struct {
 	Status     string            `bson:"status" json:"status"` // queued|running|success|failed|skipped
 	Attempts   int               `bson:"attempts" json:"attempts"`
 	ExitCode   int               `bson:"exit_code" json:"exit_code"`
-	Stdout     string            `bson:"stdout" json:"stdout"`
-	Stderr     string            `bson:"stderr" json:"stderr"`
+	LogOffset  int64             `bson:"log_offset" json:"log_offset"`
 	OutputEnv  map[string]string `bson:"output_env" json:"output_env"`
 	StartedAt  *time.Time        `bson:"started_at,omitempty" json:"started_at,omitempty"`
 	FinishedAt *time.Time        `bson:"finished_at,omitempty" json:"finished_at,omitempty"`
@@ -73,7 +85,7 @@ type ServerRun struct {
 }
 
 type WorkflowRun struct {
-	ID          string         `bson:"_id,omitempty" json:"-"`
+	ID          bson.ObjectID  `bson:"_id,omitempty" json:"-"`
 	RunID       string         `bson:"run_id" json:"run_id"`
 	WorkflowID  string         `bson:"workflow_id" json:"workflow_id"`
 	Name        string         `bson:"name" json:"name"`

server/internal/services/dispatch.go

@@ -68,6 +68,19 @@ func DispatchRunStep(serverID, commandID string, cmd *pb.RunStepCmd) error {
 	return Dispatcher.dispatch(serverID, &pb.ServerCommand{CommandId: commandID, RunStep: cmd})
 }
 
+// DispatchCleanupWorkspace tells a server's agent to remove a run's working
+// directory. Best-effort and fire-and-forget: if the agent is gone the temp dir
+// is reclaimed by the OS on reboot anyway.
+func DispatchCleanupWorkspace(serverID, workspaceID string) {
+	if !Dispatcher.IsConnected(serverID) {
+		return
+	}
+	_ = Dispatcher.dispatch(serverID, &pb.ServerCommand{
+		CommandId:        uuid.New().String(),
+		CleanupWorkspace: &pb.CleanupWorkspaceCmd{WorkspaceId: workspaceID},
+	})
+}
+
 // KeyGenParams carries all options for a generate-key command.
 type KeyGenParams struct {
 	Label      string

server/internal/services/settings.go

@@ -100,7 +100,7 @@ func VerifySecretsReadToken(token string) bool {
 	return subtle.ConstantTimeCompare(expected, got[:]) == 1
 }
 
-func SaveSettings(alerts models.AlertSettings, email models.EmailSettings) error {
+func SaveSettings(alerts models.AlertSettings, email models.EmailSettings, retentionDays *int) error {
 	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
 	defer cancel()
 
@@ -111,14 +111,31 @@ func SaveSettings(alerts models.AlertSettings, email models.EmailSettings) error
 		email.SMTPPort = 587
 	}
 
+	set := bson.M{"alerts": alerts, "email": email}
+	if retentionDays != nil {
+		set["workflow_log_retention_days"] = *retentionDays
+	}
 	_, err := db.Col("settings").UpdateOne(ctx,
 		bson.M{},
-		bson.M{"$set": bson.M{"alerts": alerts, "email": email}},
+		bson.M{"$set": set},
 		options.UpdateOne().SetUpsert(true),
 	)
 	return err
 }
 
+// GetWorkflowLogRetentionDays returns the log retention in days: 30 when unset,
+// 0 for keep-forever, or the configured value.
+func GetWorkflowLogRetentionDays() (int, error) {
+	s, err := GetSettings()
+	if err != nil {
+		return 30, err
+	}
+	if s.WorkflowLogRetentionDays == nil {
+		return 30, nil
+	}
+	return *s.WorkflowLogRetentionDays, nil
+}
+
 func SendOfflineWebhook(webhookURL, hostname, serverID, ipAddress string) {
 	payload := map[string]any{
 		"event":      "server.offline",

server/internal/services/steplogs.go

@@ -0,0 +1,217 @@
+package services
+
+import (
+	"bytes"
+	"os"
+	"path/filepath"
+	"strings"
+	"sync"
+	"time"
+
+	"github.com/mrhid6/vantage/server/internal/db"
+	"go.mongodb.org/mongo-driver/v2/bson"
+)
+
+// WorkflowLogDir returns the base directory for workflow step logs, creating it.
+func WorkflowLogDir() string {
+	dir := os.Getenv("VANTAGE_WORKFLOW_LOG_DIR")
+	if dir == "" {
+		dir = filepath.Join("data", "workflow-logs")
+	}
+	_ = os.MkdirAll(dir, 0700)
+	return dir
+}
+
+// ServerRunLogPath is the per-server-run log file path.
+func ServerRunLogPath(runID, serverID string) string {
+	return filepath.Join(WorkflowLogDir(), runID, serverID+".log")
+}
+
+// logTS is the UTC timestamp prefix stamped on every log line. Stored in UTC
+// (RFC3339, millisecond precision); the UI renders it in the viewer's timezone.
+func logTS() string {
+	return time.Now().UTC().Format("2006-01-02T15:04:05.000") + "Z"
+}
+
+// AppendMarker writes a timestamped event line to the server-run log and returns
+// the byte offset at which the write began (used as a step's log_offset).
+func AppendMarker(runID, serverID, text string) (int64, error) {
+	path := ServerRunLogPath(runID, serverID)
+	if err := os.MkdirAll(filepath.Dir(path), 0700); err != nil {
+		return 0, err
+	}
+	f, err := os.OpenFile(path, os.O_APPEND|os.O_CREATE|os.O_WRONLY, 0600)
+	if err != nil {
+		return 0, err
+	}
+	defer f.Close()
+	off, _ := f.Seek(0, 2) // current end = offset before write
+	if _, err := f.WriteString("[" + logTS() + "] " + text + "\n"); err != nil {
+		return off, err
+	}
+	return off, nil
+}
+
+// ---- streamed chunk writer, boundary-safe secret masking ----
+
+type stepLogWriter struct {
+	mu      sync.Mutex
+	f       *os.File
+	carry   []byte // bytes of an as-yet-unterminated line
+	secrets []string
+}
+
+type stepLogRegistry struct {
+	mu      sync.Mutex
+	writers map[string]*stepLogWriter
+}
+
+var StepLogs = &stepLogRegistry{writers: make(map[string]*stepLogWriter)}
+
+// Open opens (append) the server-run file for a step's streamed chunks.
+func (r *stepLogRegistry) Open(commandID, path string, secrets []string) error {
+	if err := os.MkdirAll(filepath.Dir(path), 0700); err != nil {
+		return err
+	}
+	f, err := os.OpenFile(path, os.O_APPEND|os.O_CREATE|os.O_WRONLY, 0600)
+	if err != nil {
+		return err
+	}
+	w := &stepLogWriter{f: f, secrets: secrets}
+	r.mu.Lock()
+	r.writers[commandID] = w
+	r.mu.Unlock()
+	return nil
+}
+
+func (r *stepLogRegistry) get(commandID string) *stepLogWriter {
+	r.mu.Lock()
+	defer r.mu.Unlock()
+	return r.writers[commandID]
+}
+
+// Append buffers chunks into whole lines, then writes each complete line with a
+// UTC timestamp prefix and secret masking applied. Buffering by line means a
+// secret split across a chunk boundary is always masked (the whole line is
+// assembled first) and every line carries its own timestamp.
+func (r *stepLogRegistry) Append(commandID string, data []byte) {
+	w := r.get(commandID)
+	if w == nil {
+		return
+	}
+	w.mu.Lock()
+	defer w.mu.Unlock()
+	buf := append(w.carry, data...)
+	for {
+		i := bytes.IndexByte(buf, '\n')
+		if i < 0 {
+			break
+		}
+		w.writeLine(buf[:i])
+		buf = buf[i+1:]
+	}
+	w.carry = append([]byte{}, buf...)
+}
+
+// writeLine emits one masked, timestamped log line. Caller holds w.mu.
+func (w *stepLogWriter) writeLine(line []byte) {
+	masked := maskBytes(line, w.secrets)
+	_, _ = w.f.WriteString("[" + logTS() + "] ")
+	_, _ = w.f.Write(masked)
+	_, _ = w.f.WriteString("\n")
+}
+
+// Close flushes any trailing partial line and closes the file.
+func (r *stepLogRegistry) Close(commandID string) {
+	r.mu.Lock()
+	w := r.writers[commandID]
+	delete(r.writers, commandID)
+	r.mu.Unlock()
+	if w == nil {
+		return
+	}
+	w.mu.Lock()
+	defer w.mu.Unlock()
+	if len(w.carry) > 0 {
+		w.writeLine(w.carry)
+		w.carry = nil
+	}
+	_ = w.f.Close()
+}
+
+func maskBytes(b []byte, secrets []string) []byte {
+	s := string(b)
+	for _, v := range secrets {
+		if v == "" {
+			continue
+		}
+		s = strings.ReplaceAll(s, v, "***")
+	}
+	return []byte(s)
+}
+
+// ---- retention sweeper ----
+
+// StartLogSweeper sweeps expired run-log dirs hourly (and once now).
+func StartLogSweeper() {
+	go func() {
+		sweepLogs()
+		t := time.NewTicker(time.Hour)
+		defer t.Stop()
+		for range t.C {
+			sweepLogs()
+		}
+	}()
+}
+
+func sweepLogs() {
+	days := retentionDays()
+	if days <= 0 {
+		return
+	}
+	cutoff := time.Now().AddDate(0, 0, -days)
+	base := WorkflowLogDir()
+	entries, err := os.ReadDir(base)
+	if err != nil {
+		return
+	}
+	for _, e := range entries {
+		if !e.IsDir() {
+			continue
+		}
+		runID := e.Name()
+		dir := filepath.Join(base, runID)
+		if runExpired(runID, dir, cutoff) {
+			_ = os.RemoveAll(dir)
+		}
+	}
+}
+
+// runExpired is true when the run finished before cutoff (falling back to dir
+// mtime when the run doc is gone).
+func runExpired(runID, dir string, cutoff time.Time) bool {
+	ctx, cancel := wfCtx()
+	defer cancel()
+	var run struct {
+		FinishedAt *time.Time `bson:"finished_at"`
+	}
+	err := db.Col("workflow_runs").FindOne(ctx, bson.M{"run_id": runID}).Decode(&run)
+	if err == nil {
+		if run.FinishedAt == nil {
+			return false // still running / never finished — keep
+		}
+		return run.FinishedAt.Before(cutoff)
+	}
+	// run doc gone: use dir mtime
+	if fi, e := os.Stat(dir); e == nil {
+		return fi.ModTime().Before(cutoff)
+	}
+	return false
+}
+
+func retentionDays() int {
+	if v, err := GetWorkflowLogRetentionDays(); err == nil {
+		return v
+	}
+	return 30
+}

server/internal/services/workflow_runner.go

@@ -86,6 +86,16 @@ func resolveSteps(wf *models.Workflow) ([]models.ResolvedStep, error) {
 		if err != nil {
 			return nil, err
 		}
+		inputs := map[string]string{}
+		for _, p := range lib.DeclaredInputs {
+			if ref.Inputs != nil {
+				if v, ok := ref.Inputs[p.Name]; ok {
+					inputs[p.Name] = v
+					continue
+				}
+			}
+			inputs[p.Name] = p.Default
+		}
 		rs := models.ResolvedStep{
 			Order:       ref.Order,
 			Name:        lib.Name,
@@ -94,6 +104,7 @@ func resolveSteps(wf *models.Workflow) ([]models.ResolvedStep, error) {
 			SecretRefs:  lib.SecretRefs,
 			OnFailure:   ref.OnFailure,
 			MaxRetries:  ref.MaxRetries,
+			Inputs:      inputs,
 		}
 		if ref.Overrides != nil {
 			if ref.Overrides.Script != nil {
@@ -151,16 +162,20 @@ func runServer(runID string, srvIdx int, steps []models.ResolvedStep, serverID s
 
 	if !Dispatcher.IsConnected(serverID) {
 		fin := time.Now()
+		_, _ = AppendMarker(runID, serverID, "agent not connected — server skipped")
 		setServerRun(runID, srvIdx, bson.M{"server_runs.$.status": "skipped", "server_runs.$.finished_at": fin})
 		return
 	}
 
+	_, _ = AppendMarker(runID, serverID, fmt.Sprintf("run started on %s — %d step(s), workspace vantage-run-%s", serverID, len(steps), runID))
+
 	runEnv := map[string]string{}
 	allSecrets := map[string]string{}
 	serverFailed := false
 
 	for i, step := range steps {
 		startStep(runID, serverID, i, "running")
+		stepStart := time.Now()
 		var res *pb.StepResult
 		attempts := 0
 		maxAttempts := 1
@@ -174,6 +189,9 @@ func runServer(runID string, srvIdx int, steps []models.ResolvedStep, serverID s
 			allSecrets[k] = v
 		}
 		cmdEnv := map[string]string{}
+		for k, v := range step.Inputs {
+			cmdEnv[k] = v
+		}
 		for k, v := range runEnv {
 			cmdEnv[k] = v
 		}
@@ -181,60 +199,83 @@ func runServer(runID string, srvIdx int, steps []models.ResolvedStep, serverID s
 			cmdEnv[k] = v
 		}
 
+		// Write the step marker to the server-run log and remember the offset so
+		// the UI can slice this step's output later.
+		marker := fmt.Sprintf("===== step %d/%d: %s (%s) =====", step.Order+1, len(steps), step.Name, step.Interpreter)
+		offset, _ := AppendMarker(runID, serverID, marker)
+		logPath := ServerRunLogPath(runID, serverID)
+		secretsSlice := secretValues(secretVals)
+
+		commandID := uuid.New().String()
 		for attempts < maxAttempts {
 			attempts++
-			res = dispatchAndWait(serverID, &pb.RunStepCmd{
+			if attempts > 1 {
+				_, _ = AppendMarker(runID, serverID, fmt.Sprintf("retry %d/%d after failure", attempts-1, maxAttempts-1))
+			}
+			// Open a fresh writer per attempt; the agent's eof closes it, and the
+			// defensive Close below covers a missing result.
+			_ = StepLogs.Open(commandID, logPath, secretsSlice)
+			res = dispatchAndWait(serverID, commandID, &pb.RunStepCmd{
 				Interpreter:    step.Interpreter,
 				Script:         step.Script,
 				Env:            cmdEnv,
 				TimeoutSeconds: 0,
+				WorkspaceId:    runID,
 			})
+			StepLogs.Close(commandID) // idempotent; no-op if eof already closed it
 			if res != nil && res.ExitCode == 0 {
 				break
 			}
 		}
 
-		// Mask secret values before persisting.
-		stdout, stderr := "", ""
 		exit := 1
-		outEnv := map[string]string{}    // masked copy, safe to persist
+		outEnv := map[string]string{} // masked copy, safe to persist
 		if res != nil {
-			stdout = maskSecrets(res.Stdout, allSecrets)
-			stderr = maskSecrets(res.Stderr, allSecrets)
 			exit = res.ExitCode
 			for k, v := range res.OutputEnv {
 				runEnv[k] = v // real, unmasked value threads forward to later steps
 				outEnv[k] = maskSecrets(v, allSecrets)
 			}
 		} else {
-			stderr = "[vantage] agent did not return a result"
+			_, _ = AppendMarker(runID, serverID, "agent did not return a result")
 		}
 
 		status := "success"
 		if exit != 0 {
 			status = "failed"
 		}
-		finishStep(runID, serverID, i, status, attempts, exit, stdout, stderr, outEnv)
+		finishStep(runID, serverID, i, status, attempts, exit, offset, outEnv)
+
+		dur := time.Since(stepStart).Round(time.Millisecond)
+		_, _ = AppendMarker(runID, serverID, fmt.Sprintf("step %d/%d %s — exit %d, %d attempt(s), %s",
+			step.Order+1, len(steps), status, exit, attempts, dur))
 
 		if exit != 0 {
 			switch step.OnFailure {
 			case "continue":
-				// keep going
+				_, _ = AppendMarker(runID, serverID, "on_failure=continue — proceeding to next step")
 			default: // "stop" or exhausted "retry"
 				serverFailed = true
 			}
 			if serverFailed {
+				_, _ = AppendMarker(runID, serverID, "stopping run — remaining steps skipped")
 				markRemainingSkipped(runID, serverID, i+1)
 				break
 			}
 		}
 	}
 
+	// Tell the agent to remove the run's working directory now that its steps are
+	// done (success or failure). Best-effort; the OS reclaims temp dirs anyway.
+	DispatchCleanupWorkspace(serverID, runID)
+
 	fin := time.Now()
 	status := "success"
 	if serverFailed {
 		status = "failed"
 	}
+	_, _ = AppendMarker(runID, serverID, fmt.Sprintf("run %s in %s — workspace removed",
+		status, fin.Sub(now).Round(time.Millisecond)))
 	// Persist only a masked copy of runEnv; the real (unmasked) runEnv was already
 	// used above to build cmdEnv for each step and must never be written to the DB.
 	maskedRunEnv := make(map[string]string, len(runEnv))
@@ -250,8 +291,7 @@ func runServer(runID string, srvIdx int, steps []models.ResolvedStep, serverID s
 
 // dispatchAndWait registers a waiter, dispatches the step, and blocks for the
 // result or a timeout.
-func dispatchAndWait(serverID string, cmd *pb.RunStepCmd) *pb.StepResult {
-	commandID := uuid.New().String()
+func dispatchAndWait(serverID, commandID string, cmd *pb.RunStepCmd) *pb.StepResult {
 	ch := StepResults.Await(commandID)
 	if err := DispatchRunStep(serverID, commandID, cmd); err != nil {
 		StepResults.Cancel(commandID)
@@ -322,19 +362,27 @@ func startStep(runID, serverID string, order int, status string) {
 	})
 }
 
-func finishStep(runID, serverID string, order int, status string, attempts, exit int, stdout, stderr string, outEnv map[string]string) {
+func finishStep(runID, serverID string, order int, status string, attempts, exit int, logOffset int64, outEnv map[string]string) {
 	now := time.Now()
 	updateStep(runID, serverID, order, bson.M{
 		"server_runs.$[s].steps.$[t].status":      status,
 		"server_runs.$[s].steps.$[t].attempts":    attempts,
 		"server_runs.$[s].steps.$[t].exit_code":   exit,
-		"server_runs.$[s].steps.$[t].stdout":      stdout,
-		"server_runs.$[s].steps.$[t].stderr":      stderr,
+		"server_runs.$[s].steps.$[t].log_offset":  logOffset,
 		"server_runs.$[s].steps.$[t].output_env":  outEnv,
 		"server_runs.$[s].steps.$[t].finished_at": now,
 	})
 }
 
+// secretValues returns just the values of a secret map, for masking log output.
+func secretValues(m map[string]string) []string {
+	out := make([]string, 0, len(m))
+	for _, v := range m {
+		out = append(out, v)
+	}
+	return out
+}
+
 func markRemainingSkipped(runID, serverID string, fromOrder int) {
 	ctx, cancel := wfCtx()
 	defer cancel()

server/internal/services/workflows.go

@@ -66,6 +66,9 @@ func CreateStep(s models.WorkflowStep) (*models.WorkflowStep, error) {
 	if s.SecretRefs == nil {
 		s.SecretRefs = []string{}
 	}
+	if s.DeclaredInputs == nil {
+		s.DeclaredInputs = []models.InputParam{}
+	}
 	if _, err := db.Col("workflow_steps").InsertOne(ctx, s); err != nil {
 		return nil, err
 	}
@@ -81,6 +84,7 @@ func UpdateStep(stepID string, s models.WorkflowStep) error {
 		"interpreter":      s.Interpreter,
 		"script":           s.Script,
 		"declared_outputs": s.DeclaredOutputs,
+		"declared_inputs":  s.DeclaredInputs,
 		"secret_refs":      s.SecretRefs,
 		"updated_at":       time.Now(),
 	}})
@@ -90,8 +94,38 @@ func UpdateStep(stepID string, s models.WorkflowStep) error {
 func DeleteStep(stepID string) error {
 	ctx, cancel := wfCtx()
 	defer cancel()
-	_, err := db.Col("workflow_steps").DeleteOne(ctx, bson.M{"step_id": stepID})
-	return err
+	if _, err := db.Col("workflow_steps").DeleteOne(ctx, bson.M{"step_id": stepID}); err != nil {
+		return err
+	}
+	// Cascade: remove this step from every workflow that references it, re-sequencing orders.
+	cur, err := db.Col("workflows").Find(ctx, bson.M{"steps.step_id": stepID})
+	if err != nil {
+		return err
+	}
+	defer cur.Close(ctx)
+	var wfs []models.Workflow
+	if err := cur.All(ctx, &wfs); err != nil {
+		return err
+	}
+	for _, w := range wfs {
+		kept := make([]models.WorkflowStepRef, 0, len(w.Steps))
+		for _, ref := range w.Steps {
+			if ref.StepID == stepID {
+				continue
+			}
+			kept = append(kept, ref)
+		}
+		for i := range kept {
+			kept[i].Order = i
+		}
+		if _, err := db.Col("workflows").UpdateOne(ctx,
+			bson.M{"workflow_id": w.WorkflowID},
+			bson.M{"$set": bson.M{"steps": kept, "updated_at": time.Now()}},
+		); err != nil {
+			return err
+		}
+	}
+	return nil
 }
 
 func getStep(ctx context.Context, stepID string) (*models.WorkflowStep, error) {

web/app/globals.css

@@ -36,3 +36,22 @@ body {
 ::-webkit-scrollbar-thumb:hover {
   background: #3e4160;
 }
+
+@keyframes led-pulse {
+  0%, 100% { opacity: 1; }
+  50% { opacity: 0.5; }
+}
+.led-pulse { animation: led-pulse 1.4s ease-in-out infinite; }
+
+@keyframes cell-ring {
+  0%, 100% { box-shadow: 0 0 0 0 rgba(99, 102, 241, 0.5); }
+  50% { box-shadow: 0 0 0 4px rgba(99, 102, 241, 0); }
+}
+.cell-ring { animation: cell-ring 1.4s ease-in-out infinite; }
+
+@keyframes caret-blink { 50% { opacity: 0; } }
+.caret-blink { animation: caret-blink 1s step-end infinite; }
+
+@media (prefers-reduced-motion: reduce) {
+  .led-pulse, .cell-ring, .caret-blink { animation: none; }
+}

web/app/settings/page.tsx

@@ -168,6 +168,9 @@ export default function SettingsPage() {
   const [toAddrs, setToAddrs] = useState(""); // comma-separated in UI
   const [useTLS, setUseTLS] = useState(false);
 
+  // Workflow log retention (days). 0 = keep forever.
+  const [logRetentionDays, setLogRetentionDays] = useState(30);
+
   const [saved, setSaved] = useState(false);
 
   useEffect(() => {
@@ -183,11 +186,15 @@ export default function SettingsPage() {
     setFromAddr(settings.email?.from_addr ?? "");
     setToAddrs((settings.email?.to_addrs ?? []).join(", "));
     setUseTLS(settings.email?.use_tls ?? false);
+    setLogRetentionDays(settings.workflow_log_retention_days ?? 30);
   }, [settings]);
 
   const { mutate: save, isPending } = useMutation({
-    mutationFn: (payload: { alerts: AlertSettings; email: EmailSettings }) =>
-      api.saveSettings(payload),
+    mutationFn: (payload: {
+      alerts: AlertSettings;
+      email: EmailSettings;
+      workflow_log_retention_days?: number | null;
+    }) => api.saveSettings(payload),
     onSuccess: () => {
       queryClient.invalidateQueries({ queryKey: ["settings"] });
       setSaved(true);
@@ -218,6 +225,7 @@ export default function SettingsPage() {
         to_addrs: toList,
         use_tls: useTLS,
       },
+      workflow_log_retention_days: logRetentionDays,
     });
   }
 
@@ -374,6 +382,29 @@ export default function SettingsPage() {
           </div>
         </Card>
 
+        {/* Workflow logs */}
+        <Card>
+          <CardHeader>
+            <CardTitle>Workflow Logs</CardTitle>
+          </CardHeader>
+          <p className="mb-5 text-sm text-text-secondary">
+            How long to keep workflow run logs on the server before they are
+            automatically deleted.
+          </p>
+          <Field
+            label="Log retention (days)"
+            hint="0 = keep forever. Applies to per-run step output logs."
+          >
+            <input
+              type="number"
+              min={0}
+              value={logRetentionDays}
+              onChange={(e) => setLogRetentionDays(Number(e.target.value))}
+              className="w-32 rounded-lg border border-border bg-surface-2 px-3 py-2 text-sm text-text-primary focus:border-accent/50 focus:outline-none focus:ring-1 focus:ring-accent/30"
+            />
+          </Field>
+        </Card>
+
         <div className="flex items-center gap-3">
           <Button type="submit" variant="primary" loading={isPending}>
             {saved ? "Saved!" : "Save Settings"}

web/app/workflows/[id]/page.tsx

@@ -1,470 +1,560 @@
 "use client";
 
 import { useEffect, useState } from "react";
+import Link from "next/link";
 import { useParams, useRouter } from "next/navigation";
 import { useQuery, useQueryClient } from "@tanstack/react-query";
-import {
-  api,
-  Workflow,
-  WorkflowStep,
-  WorkflowStepRef,
-  SecretGroupSummary,
-} from "@/lib/api";
-import { Button, Card } from "@/components/ui";
+import { api, Workflow, WorkflowStep, WorkflowStepRef, SecretGroupSummary } from "@/lib/api";
+import { Button } from "@/components/ui";
+import { EditStepModal } from "@/components/workflows/EditStepModal";
+import { EditWorkflowModal } from "@/components/workflows/EditWorkflowModal";
 
 const inputClass =
-  "w-full rounded-lg border border-border bg-surface-2 px-3 py-2 text-sm text-text-primary placeholder-text-secondary/50 focus:border-accent focus:outline-none focus:ring-1 focus:ring-accent";
+    "w-full rounded-lg border border-border bg-surface-2 px-3 py-2 text-sm text-text-primary placeholder-text-secondary/50 focus:border-signal focus:outline-none focus:ring-1 focus:ring-signal";
 
-function NewStepForm({ onClose }: { onClose: () => void }) {
-  const queryClient = useQueryClient();
-  const [name, setName] = useState("");
-  const [interpreter, setInterpreter] = useState<"bash" | "powershell">("bash");
-  const [script, setScript] = useState("");
-  const [saving, setSaving] = useState(false);
-  const [error, setError] = useState<string | null>(null);
+type DragPayload = { kind: "lib"; stepId: string } | { kind: "move"; from: number };
 
-  const create = async () => {
-    setSaving(true);
-    setError(null);
-    try {
-      await api.createStep({
-        name: name.trim(),
-        description: "",
-        interpreter,
-        script,
-        declared_outputs: [],
-        secret_refs: [],
-      });
-      queryClient.invalidateQueries({ queryKey: ["steps"] });
-      onClose();
-    } catch (e) {
-      setError((e as Error).message);
-    } finally {
-      setSaving(false);
-    }
-  };
-
-  return (
-    <div className="mb-3 space-y-2 rounded-lg border border-border bg-surface-2 p-2">
-      {error && <div className="text-xs text-danger">{error}</div>}
-      <input
-        className={inputClass}
-        placeholder="Step name"
-        value={name}
-        onChange={(e) => setName(e.target.value)}
-      />
-      <select
-        className={inputClass}
-        value={interpreter}
-        onChange={(e) => setInterpreter(e.target.value as "bash" | "powershell")}
-      >
-        <option value="bash">bash</option>
-        <option value="powershell">powershell</option>
-      </select>
-      <textarea
-        className={`${inputClass} h-20 font-mono text-xs`}
-        placeholder="script"
-        value={script}
-        onChange={(e) => setScript(e.target.value)}
-      />
-      <div className="flex justify-end gap-2">
-        <Button variant="ghost" size="sm" onClick={onClose}>
-          Cancel
-        </Button>
-        <Button
-          variant="primary"
-          size="sm"
-          loading={saving}
-          disabled={!name.trim()}
-          onClick={create}
+function ShellBadge({ interpreter }: { interpreter: "bash" | "powershell" }) {
+    const isBash = interpreter === "bash";
+    return (
+        <span
+            className={`rounded px-1.5 py-0.5 font-mono text-[10px] uppercase ${
+                isBash ? "bg-bash/15 text-bash" : "bg-pwsh/15 text-pwsh"
+            }`}
         >
-          Add
-        </Button>
-      </div>
-    </div>
-  );
+            {isBash ? "bash" : "pwsh"}
+        </span>
+    );
 }
 
 export default function WorkflowBuilder() {
-  const params = useParams<{ id: string }>();
-  const id = params.id;
-  const router = useRouter();
-  const [wf, setWf] = useState<Workflow | null>(null);
-  const [selected, setSelected] = useState<number | null>(null);
-  const [showNewStep, setShowNewStep] = useState(false);
-  const [saving, setSaving] = useState(false);
-  const [running, setRunning] = useState(false);
-  const [error, setError] = useState<string | null>(null);
-  const [groupKeys, setGroupKeys] = useState<Record<string, string[]>>({});
+    const params = useParams<{ id: string }>();
+    const id = params.id;
+    const router = useRouter();
+    const queryClient = useQueryClient();
 
-  const { data: loaded } = useQuery({
-    queryKey: ["workflow", id],
-    queryFn: () => api.getWorkflow(id),
-  });
-  const { data: library } = useQuery({ queryKey: ["steps"], queryFn: api.listSteps });
-  const { data: servers } = useQuery({ queryKey: ["servers"], queryFn: api.listServers });
-  const { data: secretGroups } = useQuery({
-    queryKey: ["secret-groups"],
-    queryFn: api.listSecretGroups,
-  });
+    const [wf, setWf] = useState<Workflow | null>(null);
+    const [selected, setSelected] = useState<number | null>(null);
+    const [search, setSearch] = useState("");
+    const [saving, setSaving] = useState(false);
+    const [running, setRunning] = useState(false);
+    const [error, setError] = useState<string | null>(null);
+    const [groupKeys, setGroupKeys] = useState<Record<string, string[]>>({});
+    const [editWorkflowOpen, setEditWorkflowOpen] = useState(false);
+    const [editingStep, setEditingStep] = useState<WorkflowStep | null>(null);
+    const [editStepOpen, setEditStepOpen] = useState(false);
+    const [dragOverZone, setDragOverZone] = useState<number | null>(null);
 
-  useEffect(() => {
-    if (loaded && !wf) setWf(loaded);
-  }, [loaded, wf]);
+    const { data: loaded } = useQuery({
+        queryKey: ["workflow", id],
+        queryFn: () => api.getWorkflow(id),
+    });
+    const { data: library } = useQuery({ queryKey: ["steps"], queryFn: api.listSteps });
+    const { data: secretGroups } = useQuery({
+        queryKey: ["secret-groups"],
+        queryFn: api.listSecretGroups,
+    });
 
-  // Lazily fetch the keys for every secret group so the inspector's
-  // secret-ref multiselect can offer "group/KEY" options.
-  useEffect(() => {
-    if (!secretGroups) return;
-    secretGroups.forEach((g: SecretGroupSummary) => {
-      if (groupKeys[g.group] !== undefined) return;
-      api
-        .getSecretGroup(g.group)
-        .then((res) =>
-          setGroupKeys((prev) => ({
-            ...prev,
-            [g.group]: res.secrets.map((s) => s.key),
-          }))
-        )
-        .catch(() => {
-          setGroupKeys((prev) => ({ ...prev, [g.group]: [] }));
+    useEffect(() => {
+        if (loaded && !wf) setWf(loaded);
+        // eslint-disable-next-line react-hooks/exhaustive-deps
+    }, [loaded]);
+
+    // Lazily fetch the keys for every secret group so the inspector's
+    // secret-ref checklist can offer "group/KEY" options.
+    useEffect(() => {
+        if (!secretGroups) return;
+        secretGroups.forEach((g: SecretGroupSummary) => {
+            if (groupKeys[g.group] !== undefined) return;
+            api.getSecretGroup(g.group)
+                .then((res) =>
+                    setGroupKeys((prev) => ({
+                        ...prev,
+                        [g.group]: res.secrets.map((s) => s.key),
+                    })),
+                )
+                .catch(() => {
+                    setGroupKeys((prev) => ({ ...prev, [g.group]: [] }));
+                });
         });
-    });
-    // eslint-disable-next-line react-hooks/exhaustive-deps
-  }, [secretGroups]);
+        // eslint-disable-next-line react-hooks/exhaustive-deps
+    }, [secretGroups]);
 
-  if (!wf) {
-    return <div className="p-8 text-text-secondary">Loading…</div>;
-  }
-
-  const save = async () => {
-    setSaving(true);
-    setError(null);
-    try {
-      const updated = await api.updateWorkflow(id, wf);
-      setWf(updated);
-    } catch (e) {
-      setError((e as Error).message);
-    } finally {
-      setSaving(false);
+    if (!wf) {
+        return <div className="p-8 text-text-secondary">Loading…</div>;
     }
-  };
 
-  const run = async () => {
-    setRunning(true);
-    setError(null);
-    try {
-      const { run_id } = await api.runWorkflow(id);
-      router.push(`/workflows/${id}/runs/${run_id}`);
-    } catch (e) {
-      setError((e as Error).message);
-      setRunning(false);
-    }
-  };
+    const libById = (sid: string) => library?.find((l) => l.step_id === sid);
 
-  const addStep = (s: WorkflowStep) =>
-    setWf({
-      ...wf,
-      steps: [
-        ...wf.steps,
-        { step_id: s.step_id, order: wf.steps.length, on_failure: "stop", max_retries: 0 },
-      ],
-    });
+    const sortedSteps = [...wf.steps].sort((a, b) => a.order - b.order);
+    const selectedRef = selected !== null ? sortedSteps[selected] : null;
+    const selectedLib = selectedRef ? libById(selectedRef.step_id) : null;
+    const selectedIdxInWf = selectedRef ? wf.steps.indexOf(selectedRef) : -1;
 
-  const updateRef = (idx: number, patch: Partial<WorkflowStepRef>) =>
-    setWf({
-      ...wf,
-      steps: wf.steps.map((r, i) => (i === idx ? { ...r, ...patch } : r)),
-    });
+    const save = async () => {
+        setSaving(true);
+        setError(null);
+        try {
+            const updated = await api.updateWorkflow(id, wf);
+            if (!updated || !Array.isArray(updated.steps)) {
+                setError("Save failed: server returned an unexpected response.");
+                return;
+            }
+            setWf(updated);
+        } catch (e) {
+            setError((e as Error).message);
+        } finally {
+            setSaving(false);
+        }
+    };
 
-  const removeStep = (idx: number) => {
-    const remaining = wf.steps.filter((_, i) => i !== idx).map((r, i) => ({ ...r, order: i }));
-    setWf({ ...wf, steps: remaining });
-    setSelected(null);
-  };
+    const run = async () => {
+        setRunning(true);
+        setError(null);
+        try {
+            const { run_id } = await api.runWorkflow(id);
+            router.push(`/workflows/${id}/runs/${run_id}`);
+        } catch (e) {
+            setError((e as Error).message);
+            setRunning(false);
+        }
+    };
 
-  const moveStep = (idx: number, dir: -1 | 1) => {
-    const target = idx + dir;
-    const sorted = [...wf.steps].sort((a, b) => a.order - b.order);
-    if (target < 0 || target >= sorted.length) return;
-    const next = sorted.map((r, i) => {
-      if (i === idx) return { ...r, order: sorted[target].order };
-      if (i === target) return { ...r, order: sorted[idx].order };
-      return r;
-    });
-    setWf({ ...wf, steps: next });
-    if (selected === idx) setSelected(target);
-    else if (selected === target) setSelected(idx);
-  };
+    const resequence = (steps: WorkflowStepRef[]) => steps.map((r, i) => ({ ...r, order: i }));
 
-  const toggleTargetServer = (serverId: string) => {
-    const set = new Set(wf.target_server_ids);
-    if (set.has(serverId)) set.delete(serverId);
-    else set.add(serverId);
-    setWf({ ...wf, target_server_ids: Array.from(set) });
-  };
+    const insertLibStep = (stepId: string, pos: number) => {
+        const next = [...sortedSteps];
+        next.splice(pos, 0, { step_id: stepId, order: 0, on_failure: "stop", max_retries: 0 });
+        setWf({ ...wf, steps: resequence(next) });
+    };
 
-  const libById = (sid: string) => library?.find((l) => l.step_id === sid);
+    const moveStep = (from: number, pos: number) => {
+        const next = [...sortedSteps];
+        const [item] = next.splice(from, 1);
+        const target = from < pos ? pos - 1 : pos;
+        next.splice(target, 0, item);
+        setWf({ ...wf, steps: resequence(next) });
+        if (selected === from) setSelected(target);
+        else if (selected !== null) {
+            if (from < selected && target >= selected) setSelected(selected - 1);
+            else if (from > selected && target <= selected) setSelected(selected + 1);
+        }
+    };
 
-  const sortedSteps = [...wf.steps].sort((a, b) => a.order - b.order);
-  const selectedRef = selected !== null ? sortedSteps[selected] : null;
-  const selectedLib = selectedRef ? libById(selectedRef.step_id) : null;
-  const selectedIdxInWf = selectedRef ? wf.steps.indexOf(selectedRef) : -1;
+    const handleDrop = (e: React.DragEvent, pos: number) => {
+        e.preventDefault();
+        setDragOverZone(null);
+        const raw = e.dataTransfer.getData("text/plain");
+        if (!raw) return;
+        let payload: DragPayload;
+        try {
+            payload = JSON.parse(raw);
+        } catch {
+            return;
+        }
+        if (payload.kind === "lib") {
+            insertLibStep(payload.stepId, pos);
+        } else if (payload.kind === "move") {
+            moveStep(payload.from, pos);
+        }
+    };
 
-  const toggleSecretRef = (ref: string) => {
-    if (selectedIdxInWf === -1) return;
-    const current = selectedRef?.overrides?.secret_refs ?? [];
-    const next = current.includes(ref)
-      ? current.filter((r) => r !== ref)
-      : [...current, ref];
-    updateRef(selectedIdxInWf, { overrides: { ...selectedRef?.overrides, secret_refs: next } });
-  };
+    const updateRef = (idx: number, patch: Partial<WorkflowStepRef>) =>
+        setWf({
+            ...wf,
+            steps: wf.steps.map((r, i) => (i === idx ? { ...r, ...patch } : r)),
+        });
 
-  return (
-    <div className="grid h-[calc(100vh-0px)] grid-cols-[264px_1fr_320px]">
-      {/* LEFT: library */}
-      <aside className="overflow-auto border-r border-border bg-surface p-3">
-        <div className="mb-2 flex items-center justify-between">
-          <h2 className="text-xs font-bold uppercase tracking-wide text-text-secondary">
-            Step Library
-          </h2>
-          <Button variant="ghost" size="sm" onClick={() => setShowNewStep((v) => !v)}>
-            {showNewStep ? "Close" : "Add"}
-          </Button>
-        </div>
-        {showNewStep && <NewStepForm onClose={() => setShowNewStep(false)} />}
-        {library?.map((s) => (
-          <button
-            key={s.step_id}
-            onClick={() => addStep(s)}
-            className="mb-2 block w-full rounded-lg border border-border bg-surface-2 p-2 text-left hover:border-accent"
-          >
-            <span className="font-mono text-[10px] uppercase text-accent">{s.interpreter}</span>
-            <div className="text-sm font-medium text-text-primary">{s.name}</div>
-          </button>
-        ))}
-        {library && library.length === 0 && (
-          <p className="text-xs text-text-secondary">No steps yet. Add one above.</p>
-        )}
-      </aside>
+    const removeStep = (idx: number) => {
+        const remaining = resequence(wf.steps.filter((_, i) => i !== idx));
+        setWf({ ...wf, steps: remaining });
+        setSelected(null);
+    };
 
-      {/* CENTER: canvas */}
-      <main className="overflow-auto p-6">
-        {error && (
-          <div className="mb-4 rounded-lg border border-danger/30 bg-danger/10 px-3 py-2 text-sm text-danger">
-            {error}
-          </div>
-        )}
-        <div className="mb-4 flex flex-wrap items-center gap-3">
-          <input
-            className={`${inputClass} max-w-xs`}
-            value={wf.name}
-            onChange={(e) => setWf({ ...wf, name: e.target.value })}
-          />
-          <Button variant="secondary" loading={saving} onClick={save}>
-            Save
-          </Button>
-          <Button variant="primary" loading={running} onClick={run}>
-            Run
-          </Button>
-        </div>
+    const toggleSecretRef = (ref: string) => {
+        if (selectedIdxInWf === -1 || !selectedRef) return;
+        const current = selectedRef.overrides?.secret_refs ?? [];
+        const next = current.includes(ref) ? current.filter((r) => r !== ref) : [...current, ref];
+        updateRef(selectedIdxInWf, { overrides: { ...selectedRef.overrides, secret_refs: next } });
+    };
 
-        <Card className="mb-4">
-          <h3 className="mb-2 text-xs font-bold uppercase tracking-wide text-text-secondary">
-            Target servers
-          </h3>
-          <div className="flex flex-wrap gap-2">
-            {servers?.map((s) => {
-              const isChecked = wf.target_server_ids.includes(s.server_id);
-              return (
-                <label
-                  key={s.server_id}
-                  className={`flex cursor-pointer items-center gap-2 rounded-lg border px-2 py-1 text-sm ${
-                    isChecked ? "border-accent bg-accent/10 text-text-primary" : "border-border text-text-secondary"
-                  }`}
-                >
-                  <input
-                    type="checkbox"
-                    className="accent-accent"
-                    checked={isChecked}
-                    onChange={() => toggleTargetServer(s.server_id)}
-                  />
-                  {s.hostname}
-                </label>
-              );
-            })}
-            {servers && servers.length === 0 && (
-              <p className="text-xs text-text-secondary">No servers registered.</p>
-            )}
-          </div>
-        </Card>
+    const filteredLibrary = (library ?? []).filter((s) => s.name.toLowerCase().includes(search.toLowerCase()));
+    const bashSteps = filteredLibrary.filter((s) => s.interpreter === "bash");
+    const pwshSteps = filteredLibrary.filter((s) => s.interpreter === "powershell");
 
-        <div className="mx-auto flex max-w-md flex-col items-center gap-2">
-          {sortedSteps.map((ref, i) => {
-            const lib = libById(ref.step_id);
-            const outs = sortedSteps.slice(0, i).flatMap((r) => libById(r.step_id)?.declared_outputs ?? []);
-            const script = ref.overrides?.script ?? lib?.script ?? "";
-            const wfIdx = wf.steps.indexOf(ref);
-            return (
-              <div key={i} className="w-full">
-                {i > 0 && outs.length > 0 && (
-                  <div className="mx-auto my-1 flex w-fit flex-wrap items-center justify-center gap-1 rounded-full border border-dashed border-accent/50 px-3 py-1">
-                    <span className="text-[10px] uppercase text-text-secondary">passes</span>
-                    {outs.map((o) => (
-                      <span
-                        key={o}
-                        className="rounded bg-accent px-2 py-0.5 font-mono text-[11px] text-white"
-                      >
-                        {o}
-                      </span>
-                    ))}
-                  </div>
-                )}
-                <div
-                  className={`w-full rounded-lg border bg-surface p-3 ${
-                    selected === i ? "border-accent" : "border-border"
-                  }`}
-                >
-                  <button onClick={() => setSelected(i)} className="block w-full text-left">
-                    <div className="mb-1 flex items-center gap-2">
-                      <span className="font-mono text-[10px] uppercase text-accent">
-                        {lib?.interpreter}
-                      </span>
-                      <span className="font-medium text-text-primary">
-                        {lib?.name ?? ref.step_id}
-                      </span>
-                      <span className="ml-auto rounded bg-surface-2 px-2 py-0.5 text-[10px] uppercase text-text-secondary">
-                        {ref.on_failure}
-                      </span>
-                    </div>
-                    <pre className="max-h-16 overflow-hidden text-ellipsis whitespace-pre-wrap font-mono text-[11px] text-text-secondary">
-                      {script.slice(0, 160)}
-                    </pre>
-                  </button>
-                  <div className="mt-2 flex items-center gap-2">
-                    <Button variant="ghost" size="sm" disabled={i === 0} onClick={() => moveStep(i, -1)}>
-                      ↑
+    const upstreamOutputsFor = (i: number) =>
+        Array.from(new Set(sortedSteps.slice(0, i).flatMap((r) => libById(r.step_id)?.declared_outputs ?? [])));
+
+    const DropZone = ({ pos }: { pos: number }) => (
+        <div
+            className={`h-3 w-full transition-all ${dragOverZone === pos ? "h-8 rounded bg-signal/15 border border-dashed border-signal/50" : ""}`}
+            onDragOver={(e) => {
+                e.preventDefault();
+                setDragOverZone(pos);
+            }}
+            onDragLeave={() => setDragOverZone((z) => (z === pos ? null : z))}
+            onDrop={(e) => handleDrop(e, pos)}
+        />
+    );
+
+    return (
+        <>
+            <div className="flex items-center gap-3 border-b border-border bg-surface px-4 py-3">
+                <span className="h-2.5 w-2.5 rounded-full bg-signal" />
+                <div className="flex items-center gap-1.5 text-sm">
+                    <span className="text-text-secondary">Workflows /</span>
+                    <span className="font-medium text-text-primary">{wf.name}</span>
+                    <span className="text-text-secondary">· draft</span>
+                </div>
+                <div className="ml-auto flex items-center gap-2">
+                    <span className="rounded-full border border-border bg-surface-2 px-3 py-1 text-xs text-text-secondary">
+                        {wf.target_server_ids.length} servers
+                    </span>
+                    <Link
+                        href={`/workflows/${id}/runs`}
+                        className="rounded-lg border border-border bg-surface-2 px-3 py-1.5 text-sm text-text-secondary hover:text-text-primary"
+                    >
+                        Runs
+                    </Link>
+                    <Button variant="secondary" size="sm" onClick={() => setEditWorkflowOpen(true)}>
+                        Edit
+                    </Button>
+                    <Button variant="secondary" size="sm" loading={saving} onClick={save}>
+                        Save
                     </Button>
                     <Button
-                      variant="ghost"
-                      size="sm"
-                      disabled={i === sortedSteps.length - 1}
-                      onClick={() => moveStep(i, 1)}
+                        size="sm"
+                        loading={running}
+                        onClick={run}
+                        className="bg-signal text-signal-ink border-transparent hover:bg-signal/90"
                     >
-                      ↓
+                        Run workflow
                     </Button>
-                    <Button variant="danger" size="sm" onClick={() => removeStep(wfIdx)}>
-                      Remove
-                    </Button>
-                  </div>
                 </div>
-              </div>
-            );
-          })}
-          {wf.steps.length === 0 && (
-            <p className="py-10 text-text-secondary">Click a step on the left to add it.</p>
-          )}
-        </div>
-      </main>
-
-      {/* RIGHT: inspector */}
-      <aside className="overflow-auto border-l border-border bg-surface p-4">
-        {selected === null || !selectedRef ? (
-          <p className="text-sm text-text-secondary">Select a step to configure it.</p>
-        ) : (
-          <div className="space-y-4">
-            <h2 className="text-sm font-bold text-text-primary">
-              {selectedLib?.name ?? selectedRef.step_id}
-            </h2>
-
-            <div>
-              <label className="mb-1 block text-xs uppercase text-text-secondary">Script</label>
-              <textarea
-                className={`${inputClass} h-40 font-mono text-xs`}
-                value={selectedRef.overrides?.script ?? selectedLib?.script ?? ""}
-                onChange={(e) =>
-                  updateRef(selectedIdxInWf, {
-                    overrides: { ...selectedRef.overrides, script: e.target.value },
-                  })
-                }
-              />
             </div>
 
-            <div>
-              <label className="mb-1 block text-xs uppercase text-text-secondary">On failure</label>
-              <select
-                className={inputClass}
-                value={selectedRef.on_failure}
-                onChange={(e) =>
-                  updateRef(selectedIdxInWf, {
-                    on_failure: e.target.value as WorkflowStepRef["on_failure"],
-                  })
-                }
-              >
-                <option value="stop">Stop workflow</option>
-                <option value="continue">Continue</option>
-                <option value="retry">Retry</option>
-              </select>
-            </div>
-
-            {selectedRef.on_failure === "retry" && (
-              <div>
-                <label className="mb-1 block text-xs uppercase text-text-secondary">Max retries</label>
-                <input
-                  type="number"
-                  min={1}
-                  className={inputClass}
-                  value={selectedRef.max_retries}
-                  onChange={(e) =>
-                    updateRef(selectedIdxInWf, { max_retries: parseInt(e.target.value || "0", 10) })
-                  }
-                />
-              </div>
+            {error && (
+                <div className="border-b border-danger/30 bg-danger/10 px-4 py-2 text-sm text-danger">{error}</div>
             )}
 
-            <div>
-              <label className="mb-1 block text-xs uppercase text-text-secondary">Secret refs</label>
-              <div className="max-h-56 space-y-2 overflow-auto rounded-lg border border-border p-2">
-                {secretGroups?.map((g) => (
-                  <div key={g.group}>
-                    <div className="font-mono text-[11px] font-semibold text-text-secondary">
-                      {g.group}
-                    </div>
-                    {(groupKeys[g.group] ?? []).map((key) => {
-                      const ref = `${g.group}/${key}`;
-                      const checked = (selectedRef.overrides?.secret_refs ?? []).includes(ref);
-                      return (
-                        <label
-                          key={ref}
-                          className="ml-2 flex cursor-pointer items-center gap-2 text-xs text-text-primary"
+            <div className="grid h-[calc(100vh-53px)] grid-cols-[264px_1fr_320px]">
+                {/* LEFT: library */}
+                <aside className="overflow-auto border-r border-border bg-surface p-3">
+                    <div className="mb-2 flex items-center justify-between">
+                        <h2 className="text-xs font-bold uppercase tracking-wide text-text-secondary">Step Library</h2>
+                        <Button
+                            variant="ghost"
+                            size="sm"
+                            onClick={() => {
+                                setEditingStep(null);
+                                setEditStepOpen(true);
+                            }}
                         >
-                          <input
-                            type="checkbox"
-                            className="accent-accent"
-                            checked={checked}
-                            onChange={() => toggleSecretRef(ref)}
-                          />
-                          <span className="font-mono">{key}</span>
-                        </label>
-                      );
-                    })}
-                    {(groupKeys[g.group] ?? []).length === 0 && (
-                      <p className="ml-2 text-[11px] text-text-secondary">No keys.</p>
+                            +
+                        </Button>
+                    </div>
+                    <input
+                        className={`${inputClass} mb-3`}
+                        placeholder="Search steps…"
+                        value={search}
+                        onChange={(e) => setSearch(e.target.value)}
+                    />
+
+                    {bashSteps.length > 0 && (
+                        <>
+                            <h3 className="mb-1 mt-2 text-[11px] font-bold uppercase tracking-wide text-text-secondary">
+                                Shared · Bash
+                            </h3>
+                            {bashSteps.map((s) => (
+                                <LibraryCard key={s.step_id} step={s} onAdd={() => insertLibStep(s.step_id, sortedSteps.length)} onEdit={() => {
+                                    setEditingStep(s);
+                                    setEditStepOpen(true);
+                                }} />
+                            ))}
+                        </>
                     )}
-                  </div>
-                ))}
-                {secretGroups && secretGroups.length === 0 && (
-                  <p className="text-xs text-text-secondary">No secret groups yet.</p>
-                )}
-              </div>
+
+                    {pwshSteps.length > 0 && (
+                        <>
+                            <h3 className="mb-1 mt-3 text-[11px] font-bold uppercase tracking-wide text-text-secondary">
+                                Shared · PowerShell
+                            </h3>
+                            {pwshSteps.map((s) => (
+                                <LibraryCard key={s.step_id} step={s} onAdd={() => insertLibStep(s.step_id, sortedSteps.length)} onEdit={() => {
+                                    setEditingStep(s);
+                                    setEditStepOpen(true);
+                                }} />
+                            ))}
+                        </>
+                    )}
+
+                    {filteredLibrary.length === 0 && <p className="mt-2 text-xs text-text-secondary">No steps found.</p>}
+                </aside>
+
+                {/* CENTER: canvas */}
+                <main
+                    className="overflow-auto bg-background bg-[radial-gradient(circle_at_1px_1px,theme(colors.border)_1px,transparent_0)] bg-[length:22px_22px] p-8"
+                >
+                    <div className="mx-auto flex w-[340px] flex-col items-center">
+                        <DropZone pos={0} />
+                        {sortedSteps.map((ref, i) => {
+                            const lib = libById(ref.step_id);
+                            const outs = upstreamOutputsFor(i);
+                            const script = ref.overrides?.script ?? lib?.script ?? "";
+                            const wfIdx = wf.steps.indexOf(ref);
+                            const isSelected = selected === i;
+                            return (
+                                <div key={wfIdx} className="w-full">
+                                    {i > 0 && (
+                                        <div className="flex flex-col items-center py-1">
+                                            <div className="h-[13px] w-0.5 bg-border" />
+                                            {outs.length > 0 && (
+                                                <div className="flex w-fit max-w-[300px] flex-wrap items-center justify-center gap-1 rounded-full border border-dashed border-signal/55 bg-surface px-3 py-1">
+                                                    <span className="text-[10px] uppercase text-text-secondary">passes</span>
+                                                    {outs.map((o) => (
+                                                        <span
+                                                            key={o}
+                                                            className="rounded bg-signal px-2 py-0.5 font-mono text-[11px] text-signal-ink"
+                                                        >
+                                                            {o}
+                                                        </span>
+                                                    ))}
+                                                </div>
+                                            )}
+                                            <div className="h-[13px] w-0.5 bg-border" />
+                                        </div>
+                                    )}
+                                    <div
+                                        draggable
+                                        onDragStart={(e) => {
+                                            e.dataTransfer.setData("text/plain", JSON.stringify({ kind: "move", from: i }));
+                                        }}
+                                        onClick={() => setSelected(i)}
+                                        className={`w-[340px] cursor-pointer rounded-[10px] border bg-surface p-3 ${
+                                            isSelected ? "border-signal ring-2 ring-signal/40" : "border-border"
+                                        }`}
+                                    >
+                                        <div className="mb-2 flex items-center gap-2">
+                                            <span className="grid h-5 w-5 place-items-center rounded border border-border font-mono text-[10px] text-text-secondary">
+                                                {i + 1}
+                                            </span>
+                                            <span className="text-sm font-medium text-text-primary">{lib?.name ?? ref.step_id}</span>
+                                            {lib && <ShellBadge interpreter={lib.interpreter} />}
+                                        </div>
+                                        <pre className="max-h-16 overflow-hidden text-ellipsis whitespace-pre-wrap rounded border border-border bg-surface-2 p-2 font-mono text-xs text-text-secondary">
+                                            {script.slice(0, 200)}
+                                        </pre>
+                                    </div>
+                                    <DropZone pos={i + 1} />
+                                </div>
+                            );
+                        })}
+                        {sortedSteps.length === 0 && (
+                            <button
+                                onDragOver={(e) => e.preventDefault()}
+                                onDrop={(e) => handleDrop(e, 0)}
+                                className="mt-2 w-full rounded-[10px] border border-dashed border-border bg-surface py-6 text-sm text-text-secondary hover:border-signal/50 hover:text-text-primary"
+                            >
+                                + Drop a step here
+                            </button>
+                        )}
+                    </div>
+                </main>
+
+                {/* RIGHT: inspector */}
+                <aside className="overflow-auto border-l border-border bg-surface p-4">
+                    {selected === null || !selectedRef ? (
+                        <p className="text-sm text-text-secondary">Select a step to configure it.</p>
+                    ) : (
+                        <div className="space-y-4">
+                            <div>
+                                <div className="mb-1 text-[11px] font-bold uppercase tracking-wide text-text-secondary">
+                                    Step {selected + 1} · Inspector
+                                </div>
+                                <div className="flex items-center gap-2">
+                                    {selectedLib && <ShellBadge interpreter={selectedLib.interpreter} />}
+                                    <h2 className="text-sm font-bold text-text-primary">{selectedLib?.name ?? selectedRef.step_id}</h2>
+                                </div>
+                            </div>
+
+                            <div className="border-b border-border pb-4">
+                                <label className="mb-1 block text-xs uppercase text-text-secondary">Command</label>
+                                <textarea
+                                    className={`${inputClass} h-32 font-mono text-xs`}
+                                    value={selectedRef.overrides?.script ?? selectedLib?.script ?? ""}
+                                    onChange={(e) =>
+                                        updateRef(selectedIdxInWf, {
+                                            overrides: { ...selectedRef.overrides, script: e.target.value },
+                                        })
+                                    }
+                                />
+                                <p className="mt-1 text-xs text-text-secondary">
+                                    Write <code className="text-signal">KEY=value</code> to <code className="text-signal">$WORKFLOW_ENV</code> to expose
+                                    it to later steps.
+                                </p>
+                            </div>
+
+                            {(selectedLib?.declared_inputs ?? []).length > 0 && (
+                                <div className="border-b border-border pb-4">
+                                    <label className="mb-2 block text-xs uppercase text-text-secondary">Inputs</label>
+                                    <div className="space-y-2">
+                                        {selectedLib?.declared_inputs.map((param) => (
+                                            <div key={param.name}>
+                                                <div className="mb-1 font-mono text-xs text-text-primary">{param.name}</div>
+                                                {param.description && (
+                                                    <div className="mb-1 text-[11px] text-text-secondary">{param.description}</div>
+                                                )}
+                                                <input
+                                                    className={inputClass}
+                                                    placeholder={param.default}
+                                                    value={selectedRef.inputs?.[param.name] ?? ""}
+                                                    onChange={(e) =>
+                                                        updateRef(selectedIdxInWf, {
+                                                            inputs: { ...selectedRef.inputs, [param.name]: e.target.value },
+                                                        })
+                                                    }
+                                                />
+                                            </div>
+                                        ))}
+                                    </div>
+                                </div>
+                            )}
+
+                            <div className="border-b border-border pb-4">
+                                <label className="mb-2 block text-xs uppercase text-text-secondary">Inputs · from upstream</label>
+                                <div className="flex flex-wrap gap-1">
+                                    {upstreamOutputsFor(selected).length === 0 && (
+                                        <p className="text-xs text-text-secondary">No upstream outputs.</p>
+                                    )}
+                                    {upstreamOutputsFor(selected).map((o) => (
+                                        <span key={o} className="flex items-center gap-1 rounded bg-surface-2 border border-border px-2 py-0.5 font-mono text-[11px] text-text-primary">
+                                            <span className="text-[9px] uppercase text-text-secondary">in</span>
+                                            {o}
+                                        </span>
+                                    ))}
+                                </div>
+                            </div>
+
+                            <div className="border-b border-border pb-4">
+                                <label className="mb-2 block text-xs uppercase text-text-secondary">Outputs · to $WORKFLOW_ENV</label>
+                                <div className="flex flex-wrap gap-1">
+                                    {(selectedLib?.declared_outputs ?? []).length === 0 && (
+                                        <p className="text-xs text-text-secondary">No declared outputs.</p>
+                                    )}
+                                    {(selectedLib?.declared_outputs ?? []).map((o) => (
+                                        <span key={o} className="flex items-center gap-1 rounded bg-signal px-2 py-0.5 font-mono text-[11px] text-signal-ink">
+                                            <span className="text-[9px] uppercase">out</span>
+                                            {o}
+                                        </span>
+                                    ))}
+                                </div>
+                            </div>
+
+                            <div className="border-b border-border pb-4">
+                                <label className="mb-2 block text-xs uppercase text-text-secondary">Secret refs</label>
+                                <div className="max-h-56 space-y-2 overflow-auto rounded-lg border border-border p-2">
+                                    {secretGroups?.map((g) => (
+                                        <div key={g.group}>
+                                            <div className="font-mono text-[11px] font-semibold text-text-secondary">{g.group}</div>
+                                            {(groupKeys[g.group] ?? []).map((key) => {
+                                                const ref = `${g.group}/${key}`;
+                                                const checked = (selectedRef.overrides?.secret_refs ?? []).includes(ref);
+                                                return (
+                                                    <label key={ref} className="ml-2 flex cursor-pointer items-center gap-2 text-xs text-text-primary">
+                                                        <input
+                                                            type="checkbox"
+                                                            className="accent-signal"
+                                                            checked={checked}
+                                                            onChange={() => toggleSecretRef(ref)}
+                                                        />
+                                                        <span className="font-mono">{key}</span>
+                                                    </label>
+                                                );
+                                            })}
+                                            {(groupKeys[g.group] ?? []).length === 0 && <p className="ml-2 text-[11px] text-text-secondary">No keys.</p>}
+                                        </div>
+                                    ))}
+                                    {secretGroups && secretGroups.length === 0 && <p className="text-xs text-text-secondary">No secret groups yet.</p>}
+                                </div>
+                            </div>
+
+                            <div className="border-b border-border pb-4">
+                                <label className="mb-1 block text-xs uppercase text-text-secondary">On failure</label>
+                                <select
+                                    className={inputClass}
+                                    value={selectedRef.on_failure}
+                                    onChange={(e) =>
+                                        updateRef(selectedIdxInWf, {
+                                            on_failure: e.target.value as WorkflowStepRef["on_failure"],
+                                        })
+                                    }
+                                >
+                                    <option value="stop">Stop workflow</option>
+                                    <option value="continue">Continue</option>
+                                    <option value="retry">Retry</option>
+                                </select>
+                                {selectedRef.on_failure === "retry" && (
+                                    <div className="mt-2">
+                                        <label className="mb-1 block text-xs uppercase text-text-secondary">Max retries</label>
+                                        <input
+                                            type="number"
+                                            min={1}
+                                            className={inputClass}
+                                            value={selectedRef.max_retries}
+                                            onChange={(e) => updateRef(selectedIdxInWf, { max_retries: parseInt(e.target.value || "0", 10) })}
+                                        />
+                                    </div>
+                                )}
+                            </div>
+
+                            <Button variant="danger" size="sm" onClick={() => removeStep(selectedIdxInWf)}>
+                                Remove from workflow
+                            </Button>
+                        </div>
+                    )}
+                </aside>
             </div>
 
-            <Button variant="danger" size="sm" onClick={() => removeStep(selectedIdxInWf)}>
-              Remove step
-            </Button>
-          </div>
-        )}
-      </aside>
-    </div>
-  );
+            <EditWorkflowModal open={editWorkflowOpen} workflow={wf} onSaved={(w) => setWf(w)} onClose={() => setEditWorkflowOpen(false)} />
+            <EditStepModal
+                key={editingStep?.step_id ?? "new"}
+                open={editStepOpen}
+                step={editingStep}
+                onClose={() => {
+                    setEditStepOpen(false);
+                    queryClient.invalidateQueries({ queryKey: ["steps"] });
+                }}
+            />
+        </>
+    );
+}
+
+function LibraryCard({ step, onAdd, onEdit }: { step: WorkflowStep; onAdd: () => void; onEdit: () => void }) {
+    return (
+        <div
+            draggable
+            onDragStart={(e) => {
+                e.dataTransfer.setData("text/plain", JSON.stringify({ kind: "lib", stepId: step.step_id }));
+            }}
+            onClick={onAdd}
+            className="group relative mb-2 cursor-grab rounded-lg border border-border bg-surface-2 p-2 text-left hover:border-signal/50"
+        >
+            <div className="mb-1 flex items-center gap-2">
+                <span className="text-text-secondary">⠿</span>
+                <ShellBadge interpreter={step.interpreter} />
+                <span className="text-sm font-medium text-text-primary">{step.name}</span>
+                <button
+                    onClick={(e) => {
+                        e.stopPropagation();
+                        onEdit();
+                    }}
+                    className="ml-auto hidden text-text-secondary hover:text-text-primary group-hover:block"
+                    title="Edit step"
+                >
+                    ✎
+                </button>
+            </div>
+            {step.description && <p className="text-xs text-text-secondary">{step.description}</p>}
+        </div>
+    );
 }

web/app/workflows/[id]/runs/[runId]/page.tsx

@@ -1,102 +1,442 @@
 "use client";
 
 import { useParams } from "next/navigation";
+import { useEffect, useMemo, useRef, useState } from "react";
 import { useQuery, useQueryClient } from "@tanstack/react-query";
-import { api, ServerRun, StepRun } from "@/lib/api";
-import { Button, Badge, Card } from "@/components/ui";
+import { api, ServerRun, StepRun, WorkflowRun } from "@/lib/api";
+import { Button } from "@/components/ui";
 
-type BadgeVariant = "success" | "warning" | "danger" | "neutral" | "accent";
+// ---- status vocabulary ----------------------------------------------------
 
-const statusVariant: Record<string, BadgeVariant> = {
-  success: "success",
-  failed: "danger",
-  running: "accent",
-  queued: "neutral",
-  skipped: "neutral",
-  cancelled: "warning",
+type CellKind = "done" | "fail" | "run" | "wait" | "skip" | "warn";
+
+function cellKind(status: string): CellKind {
+    switch (status) {
+        case "success":
+            return "done";
+        case "failed":
+            return "fail";
+        case "running":
+            return "run";
+        case "skipped":
+            return "skip";
+        case "cancelled":
+            return "warn";
+        default:
+            return "wait"; // queued / pending / missing
+    }
+}
+
+const cellGlyph: Record<CellKind, string> = {
+    done: "✓",
+    fail: "✕",
+    run: "●",
+    wait: "○",
+    skip: "–",
+    warn: "!",
 };
 
-function StatusBadge({ status }: { status: string }) {
-  return <Badge variant={statusVariant[status] ?? "neutral"}>{status}</Badge>;
+const cellClass: Record<CellKind, string> = {
+    done: "bg-success/15 text-success",
+    fail: "bg-danger/15 text-danger",
+    run: "bg-accent/15 text-accent",
+    wait: "text-border",
+    skip: "text-text-secondary",
+    warn: "bg-warning/15 text-warning",
+};
+
+// ---- run-level status pill ------------------------------------------------
+
+type PillKind = "running" | "success" | "failed" | "neutral";
+
+function pillKind(status: string): PillKind {
+    if (status === "running") return "running";
+    if (status === "success") return "success";
+    if (status === "failed" || status === "cancelled") return "failed";
+    return "neutral";
+}
+
+const pillClass: Record<PillKind, string> = {
+    running: "text-accent border-accent/40 bg-accent/10",
+    success: "text-success border-success/35 bg-success/10",
+    failed: "text-danger border-danger/35 bg-danger/10",
+    neutral: "text-text-secondary border-border bg-surface-2",
+};
+
+const pillLed: Record<PillKind, string> = {
+    running: "bg-accent led-pulse",
+    success: "bg-success",
+    failed: "bg-danger",
+    neutral: "bg-text-secondary",
+};
+
+function StatusPill({ status, small }: { status: string; small?: boolean }) {
+    const kind = pillKind(status);
+    return (
+        <span
+            className={`inline-flex items-center gap-2 rounded-full border font-mono font-semibold uppercase tracking-wide ${
+                small ? "px-2 py-0.5 text-[10px]" : "px-2.5 py-1 text-xs"
+            } ${pillClass[kind]}`}
+        >
+            <span className={`h-1.5 w-1.5 rounded-full ${pillLed[kind]}`} />
+            {status}
+        </span>
+    );
+}
+
+// ---- time helpers ---------------------------------------------------------
+
+function fmtDuration(ms: number): string {
+    if (ms < 0) ms = 0;
+    const s = Math.floor(ms / 1000);
+    if (s < 60) return `${s}s`;
+    const m = Math.floor(s / 60);
+    const rem = s % 60;
+    if (m < 60) return `${m}m ${rem}s`;
+    const h = Math.floor(m / 60);
+    return `${h}h ${m % 60}m`;
+}
+
+function stepDuration(st: StepRun, running: boolean, now: number): string {
+    if (!st.started_at) return st.status === "queued" ? "queued" : "";
+    const start = new Date(st.started_at).getTime();
+    const end = st.finished_at ? new Date(st.finished_at).getTime() : running ? now : start;
+    return fmtDuration(end - start);
+}
+
+// ---- live log terminal ----------------------------------------------------
+
+function LogTerminal({ runId, server }: { runId: string; server: ServerRun }) {
+    const [text, setText] = useState("");
+    const preRef = useRef<HTMLDivElement>(null);
+    const running = server.status === "running";
+    const serverId = server.server_id;
+
+    useEffect(() => {
+        setText("");
+        if (running) {
+            const es = new EventSource(api.serverRunLogStreamUrl(runId, serverId), {
+                withCredentials: true,
+            });
+            es.onmessage = (e) => setText((t) => t + e.data + "\n");
+            es.addEventListener("done", () => es.close());
+            es.onerror = () => es.close();
+            return () => es.close();
+        }
+        api.getServerRunLog(runId, serverId)
+            .then(setText)
+            .catch(() => setText(""));
+    }, [running, runId, serverId]);
+
+    useEffect(() => {
+        preRef.current?.scrollTo(0, preRef.current.scrollHeight);
+    }, [text]);
+
+    const activeStep = server.steps.find((s) => s.status === "running") ?? [...server.steps].reverse().find((s) => s.started_at);
+
+    return (
+        <div className="overflow-hidden rounded-xl border border-border bg-[#0a0b10]">
+            <div className="flex items-center justify-between gap-2 border-b border-border bg-surface px-4 py-3">
+                <span className="truncate font-mono text-[13px] font-semibold text-text-primary">
+                    {activeStep ? activeStep.name : "Output"} <span className="font-normal text-text-secondary">{server.hostname}</span>
+                </span>
+                {running && (
+                    <span className="inline-flex items-center gap-1.5 font-mono text-[10.5px] uppercase tracking-wide text-accent">
+                        <span className="h-1.5 w-1.5 rounded-full bg-accent led-pulse" />
+                        Streaming
+                    </span>
+                )}
+            </div>
+            <div ref={preRef} className="max-h-[340px] overflow-auto whitespace-pre-wrap px-4 py-3.5 font-mono text-[12.5px] leading-relaxed text-text-secondary">
+                {text ? <LogLines text={text} /> : running ? "Waiting for output…" : "No output."}
+                {running && text && <span className="ml-0.5 inline-block h-3.5 w-[7px] translate-y-[2px] bg-accent caret-blink align-baseline" />}
+            </div>
+        </div>
+    );
+}
+
+// LogLines renders the raw server-run log, parsing each line's leading UTC
+// timestamp ([2026-07-20T12:04:02.000Z]) and rendering it in the viewer's local
+// timezone. Event markers (===== …) are highlighted so the run's shape scans.
+const TS_RE = /^\[(\d{4}-\d{2}-\d{2}T[\d:.]+Z)\]\s?(.*)$/;
+
+function LogLines({ text }: { text: string }) {
+    const lines = text.replace(/\n$/, "").split("\n");
+    return (
+        <>
+            {lines.map((line, i) => {
+                const m = TS_RE.exec(line);
+                if (!m) {
+                    return (
+                        <span key={i} className="block">
+                            {line || " "}
+                        </span>
+                    );
+                }
+                const local = new Date(m[1]).toLocaleTimeString([], { hour12: false });
+                const body = m[2];
+                const isMarker = body.startsWith("=====");
+                return (
+                    <span key={i} className="block">
+                        <span className="select-none text-[#565b74]" title={m[1]}>
+                            {local}{" "}
+                        </span>
+                        <span className={isMarker ? "font-semibold text-accent" : ""}>{body || " "}</span>
+                    </span>
+                );
+            })}
+        </>
+    );
+}
+
+// ---- step list ------------------------------------------------------------
+
+function StepList({ server, now }: { server: ServerRun; now: number }) {
+    const running = server.status === "running";
+    return (
+        <div className="overflow-hidden rounded-xl border border-border bg-surface">
+            <div className="flex items-center justify-between gap-2 border-b border-border px-4 py-3">
+                <span className="font-mono text-[13px] font-semibold text-text-primary">
+                    Steps <span className="font-normal text-text-secondary">{server.steps.length}</span>
+                </span>
+                <StatusPill status={server.status} small />
+            </div>
+            <div className="flex flex-col gap-0.5 p-1.5">
+                {server.steps.map((st) => {
+                    const kind = cellKind(st.status);
+                    return (
+                        <div
+                            key={st.order}
+                            className={`grid grid-cols-[20px_1fr_auto] items-center gap-2.5 rounded-lg px-3 py-2.5 text-[13px] hover:bg-surface-2 ${st.status === "running" ? "bg-accent/[0.06]" : ""}`}
+                        >
+                            <span className="text-right font-mono text-[11px] text-text-secondary">{String(st.order + 1).padStart(2, "0")}</span>
+                            <span className="flex items-center gap-2 font-medium text-text-primary">
+                                <span className={`font-mono ${cellClass[kind].replace(/bg-\S+/, "")}`}>{cellGlyph[kind]}</span>
+                                {st.name}
+                            </span>
+                            <span className="text-right font-mono text-[10.5px] text-text-secondary">
+                                {st.status === "failed" && <span className="text-danger">exit {st.exit_code} · </span>}
+                                {st.attempts > 1 ? `${st.attempts} tries` : "1 try"}
+                                {stepDuration(st, running, now) ? ` · ${stepDuration(st, running, now)}` : ""}
+                            </span>
+                        </div>
+                    );
+                })}
+                {server.steps.length === 0 && <p className="px-3 py-2 text-xs text-text-secondary">No steps yet.</p>}
+            </div>
+        </div>
+    );
+}
+
+// ---- execution matrix (signature) -----------------------------------------
+
+interface Column {
+    order: number;
+    name: string;
+}
+
+function buildColumns(run: WorkflowRun): Column[] {
+    const byOrder = new Map<number, string>();
+    for (const sr of run.server_runs) {
+        for (const st of sr.steps) {
+            if (!byOrder.has(st.order)) byOrder.set(st.order, st.name);
+        }
+    }
+    return [...byOrder.entries()].map(([order, name]) => ({ order, name })).sort((a, b) => a.order - b.order);
+}
+
+function ExecutionMatrix({ run, columns, selected, onSelect }: { run: WorkflowRun; columns: Column[]; selected: string; onSelect: (serverId: string) => void }) {
+    return (
+        <div className="overflow-x-auto rounded-xl border border-border bg-surface">
+            <table className="w-full border-collapse font-mono text-[12.5px]">
+                <thead>
+                    <tr>
+                        <th className="border-b border-border px-4 py-3 text-left align-bottom text-xs font-semibold uppercase tracking-wider text-text-primary">Server</th>
+                        {columns.map((c) => (
+                            <th key={c.order} className="whitespace-nowrap border-b border-border px-3.5 py-3 align-bottom text-[11px] font-medium text-text-secondary">
+                                <span className="block text-[10px] text-border">{String(c.order + 1).padStart(2, "0")}</span>
+                                {c.name}
+                            </th>
+                        ))}
+                    </tr>
+                </thead>
+                <tbody>
+                    {run.server_runs.map((sr) => {
+                        const byOrder = new Map(sr.steps.map((s) => [s.order, s]));
+                        const isSel = sr.server_id === selected;
+                        return (
+                            <tr key={sr.server_id} onClick={() => onSelect(sr.server_id)} className={`cursor-pointer ${isSel ? "bg-accent/5" : "hover:bg-white/[0.02]"}`}>
+                                <th className="min-w-[240px] border-b border-r border-border px-4 py-3 text-left font-medium text-text-primary">
+                                    <div className="flex items-center gap-3">
+                                        <span className="flex-1 whitespace-nowrap">{sr.hostname}</span>
+                                        <StatusPill status={sr.status} small />
+                                    </div>
+                                </th>
+                                {columns.map((c) => {
+                                    const st = byOrder.get(c.order);
+                                    const kind = st ? cellKind(st.status) : "wait";
+                                    return (
+                                        <td key={c.order} className="relative border-b border-r border-border last:border-r-0">
+                                            <span className="flex h-[54px] items-center justify-center">
+                                                <span className={`relative flex h-[26px] w-[26px] items-center justify-center rounded-md ${cellClass[kind]}`}>
+                                                    {kind === "run" && <span className="absolute inset-0 rounded-md border border-accent/50 cell-ring" />}
+                                                    {cellGlyph[kind]}
+                                                </span>
+                                            </span>
+                                        </td>
+                                    );
+                                })}
+                            </tr>
+                        );
+                    })}
+                </tbody>
+            </table>
+        </div>
+    );
+}
+
+// ---- page -----------------------------------------------------------------
+
+function SectionLabel({ children }: { children: React.ReactNode }) {
+    return (
+        <div className="mb-3 mt-8 flex items-center gap-2.5 font-mono text-[11px] uppercase tracking-widest text-text-secondary">
+            {children}
+            <span className="h-px flex-1 bg-border" />
+        </div>
+    );
 }
 
 export default function RunDetail() {
-  const { runId } = useParams<{ runId: string }>();
-  const queryClient = useQueryClient();
+    const { runId } = useParams<{ runId: string }>();
+    const queryClient = useQueryClient();
+    const [selected, setSelected] = useState<string | null>(null);
+    const [now, setNow] = useState(() => Date.now());
 
-  const { data: run, isLoading } = useQuery({
-    queryKey: ["run", runId],
-    queryFn: () => api.getRun(runId),
-    refetchInterval: (query) => (query.state.data?.status === "running" ? 2000 : false),
-  });
+    const { data: run, isLoading } = useQuery({
+        queryKey: ["run", runId],
+        queryFn: () => api.getRun(runId),
+        refetchInterval: (query) => (query.state.data?.status === "running" ? 2000 : false),
+    });
 
-  const cancel = async () => {
-    await api.cancelRun(runId);
-    queryClient.invalidateQueries({ queryKey: ["run", runId] });
-  };
+    const running = run?.status === "running";
 
-  if (isLoading || !run) {
-    return <div className="p-8 text-text-secondary">Loading…</div>;
-  }
+    // tick the elapsed clock while running
+    useEffect(() => {
+        if (!running) return;
+        const t = setInterval(() => setNow(Date.now()), 1000);
+        return () => clearInterval(t);
+    }, [running]);
 
-  return (
-    <div className="p-8">
-      <div className="mb-6 flex items-center justify-between">
-        <div>
-          <h1 className="text-2xl font-bold text-text-primary">{run.name}</h1>
-          <p className="mt-1 flex items-center gap-2 text-sm text-text-secondary">
-            <span>Run {run.run_id.slice(0, 8)}</span>
-            <StatusBadge status={run.status} />
-          </p>
+    const columns = useMemo(() => (run ? buildColumns(run) : []), [run]);
+
+    // default selection: first running server, else first server
+    const selectedServer = useMemo(() => {
+        if (!run || run.server_runs.length === 0) return null;
+        if (selected) {
+            const match = run.server_runs.find((s) => s.server_id === selected);
+            if (match) return match;
+        }
+        return run.server_runs.find((s) => s.status === "running") ?? run.server_runs[0];
+    }, [run, selected]);
+
+    const cancel = async () => {
+        await api.cancelRun(runId);
+        queryClient.invalidateQueries({ queryKey: ["run", runId] });
+    };
+
+    if (isLoading || !run) {
+        return <div className="p-8 text-text-secondary">Loading…</div>;
+    }
+
+    const totalSteps = run.server_runs.reduce((n, s) => n + s.steps.length, 0);
+    const doneSteps = run.server_runs.reduce((n, s) => n + s.steps.filter((st) => st.status === "success").length, 0);
+    const succeeded = run.server_runs.filter((s) => s.status === "success").length;
+    const failed = run.server_runs.filter((s) => s.status === "failed" || s.status === "cancelled").length;
+
+    const startMs = run.started_at ? new Date(run.started_at).getTime() : now;
+    const endMs = run.finished_at ? new Date(run.finished_at).getTime() : now;
+    const elapsed = fmtDuration(endMs - startMs);
+    const ago = fmtDuration(now - startMs);
+
+    return (
+        <div className="mx-auto max-w-[1180px] p-8 pb-16">
+            {/* identity bar */}
+            <div className="flex flex-wrap items-start justify-between gap-6">
+                <div>
+                    <div className="mb-2 font-mono text-xs uppercase tracking-wide text-text-secondary">Workflows / {run.name} / Runs</div>
+                    <h1 className="text-[28px] font-semibold tracking-tight text-text-primary">{run.name}</h1>
+                    <div className="mt-2.5 flex flex-wrap items-center gap-x-4 gap-y-1 font-mono text-[12.5px] text-text-secondary">
+                        <span>
+                            run <b className="font-medium text-text-primary">{run.run_id.slice(0, 8)}</b>
+                        </span>
+                        <span className="h-[3px] w-[3px] rounded-full bg-border" />
+                        <span>
+                            triggered by <b className="font-medium text-text-primary">{run.triggered_by || "—"}</b>
+                        </span>
+                        <span className="h-[3px] w-[3px] rounded-full bg-border" />
+                        <span>
+                            started <b className="font-medium text-text-primary">{ago}</b> ago
+                        </span>
+                        <span className="h-[3px] w-[3px] rounded-full bg-border" />
+                        <span>
+                            elapsed <b className="font-medium text-text-primary">{elapsed}</b>
+                        </span>
+                    </div>
+                </div>
+                <div className="flex items-center gap-3.5">
+                    <StatusPill status={run.status} />
+                    {running && (
+                        <Button variant="danger" onClick={cancel}>
+                            Cancel run
+                        </Button>
+                    )}
+                </div>
+            </div>
+
+            {/* summary strip */}
+            <div className="mt-6 grid grid-cols-2 gap-px overflow-hidden rounded-xl border border-border bg-border sm:grid-cols-4">
+                <div className="bg-surface px-[18px] py-4">
+                    <div className="font-mono text-[10.5px] uppercase tracking-wider text-text-secondary">Servers</div>
+                    <div className="mt-1 font-mono text-[22px] font-semibold tabular-nums text-text-primary">{run.server_runs.length}</div>
+                </div>
+                <div className="bg-surface px-[18px] py-4">
+                    <div className="font-mono text-[10.5px] uppercase tracking-wider text-text-secondary">Succeeded</div>
+                    <div className="mt-1 font-mono text-[22px] font-semibold tabular-nums text-success">
+                        {succeeded}
+                        <small className="text-sm font-medium text-text-secondary"> / {run.server_runs.length}</small>
+                    </div>
+                </div>
+                <div className="bg-surface px-[18px] py-4">
+                    <div className="font-mono text-[10.5px] uppercase tracking-wider text-text-secondary">Failed</div>
+                    <div className={`mt-1 font-mono text-[22px] font-semibold tabular-nums ${failed > 0 ? "text-danger" : "text-text-primary"}`}>{failed}</div>
+                </div>
+                <div className="bg-surface px-[18px] py-4">
+                    <div className="font-mono text-[10.5px] uppercase tracking-wider text-text-secondary">Steps done</div>
+                    <div className="mt-1 font-mono text-[22px] font-semibold tabular-nums text-text-primary">
+                        {doneSteps}
+                        <small className="text-sm font-medium text-text-secondary"> / {totalSteps}</small>
+                    </div>
+                </div>
+            </div>
+
+            {run.server_runs.length === 0 ? (
+                <p className="mt-8 text-text-secondary">No servers targeted by this run.</p>
+            ) : (
+                <>
+                    <SectionLabel>Execution matrix</SectionLabel>
+                    <ExecutionMatrix run={run} columns={columns} selected={selectedServer?.server_id ?? ""} onSelect={setSelected} />
+
+                    {selectedServer && (
+                        <>
+                            <SectionLabel>{selectedServer.hostname}&nbsp;·&nbsp;steps &amp; live output</SectionLabel>
+                            <div className="grid grid-cols-1 items-start gap-4 md:grid-cols-[320px_1fr]">
+                                <StepList server={selectedServer} now={now} />
+                                <LogTerminal runId={run.run_id} server={selectedServer} />
+                            </div>
+                        </>
+                    )}
+                </>
+            )}
         </div>
-        {run.status === "running" && (
-          <Button variant="danger" onClick={cancel}>
-            Cancel
-          </Button>
-        )}
-      </div>
-
-      <div className="grid gap-4 md:grid-cols-2 lg:grid-cols-3">
-        {run.server_runs.map((sr: ServerRun) => (
-          <Card key={sr.server_id}>
-            <div className="mb-3 flex items-center justify-between">
-              <span className="font-medium text-text-primary">{sr.hostname}</span>
-              <StatusBadge status={sr.status} />
-            </div>
-            <div className="space-y-2">
-              {sr.steps.map((st: StepRun) => (
-                <details
-                  key={st.order}
-                  className="rounded-lg border border-border bg-surface-2 p-2"
-                >
-                  <summary className="flex cursor-pointer items-center justify-between gap-2">
-                    <span className="text-sm text-text-primary">{st.name}</span>
-                    <span className="flex items-center gap-2">
-                      <span className="text-xs text-text-secondary">
-                        attempts: {st.attempts}
-                        {st.status === "failed" ? ` · exit ${st.exit_code}` : ""}
-                      </span>
-                      <StatusBadge status={st.status} />
-                    </span>
-                  </summary>
-                  {(st.stdout || st.stderr) && (
-                    <pre className="mt-2 max-h-64 overflow-auto rounded bg-black/40 p-2 font-mono text-xs text-text-secondary">
-                      {st.stdout}
-                      {st.stderr ? `\n${st.stderr}` : ""}
-                    </pre>
-                  )}
-                </details>
-              ))}
-              {sr.steps.length === 0 && (
-                <p className="text-xs text-text-secondary">No steps yet.</p>
-              )}
-            </div>
-          </Card>
-        ))}
-        {run.server_runs.length === 0 && (
-          <p className="text-text-secondary">No servers targeted by this run.</p>
-        )}
-      </div>
-    </div>
-  );
+    );
 }

web/app/workflows/[id]/runs/page.tsx

@@ -0,0 +1,78 @@
+"use client";
+
+import Link from "next/link";
+import { useParams } from "next/navigation";
+import { useQuery } from "@tanstack/react-query";
+import { api, WorkflowRun } from "@/lib/api";
+import { Card, Table, Thead, Tbody, Tr, Th, Td, Badge } from "@/components/ui";
+
+type BadgeVariant = "success" | "warning" | "danger" | "neutral" | "accent";
+
+const statusVariant: Record<string, BadgeVariant> = {
+  success: "success",
+  failed: "danger",
+  running: "warning",
+  cancelled: "neutral",
+  queued: "neutral",
+};
+
+export default function WorkflowRunsPage() {
+  const { id } = useParams<{ id: string }>();
+  const { data: wf } = useQuery({ queryKey: ["workflow", id], queryFn: () => api.getWorkflow(id) });
+  const { data: runs, isLoading, error } = useQuery({ queryKey: ["runs", id], queryFn: () => api.listRuns(id) });
+
+  return (
+    <div className="p-8">
+      <div className="mb-6">
+        <Link href={`/workflows/${id}`} className="text-sm text-text-secondary hover:text-text-primary">
+          ← Back to builder
+        </Link>
+        <h1 className="mt-2 text-2xl font-bold text-text-primary">Runs · {wf?.name ?? ""}</h1>
+      </div>
+
+      <Card padding={false}>
+        {isLoading ? (
+          <div className="flex items-center justify-center py-20">
+            <div className="h-8 w-8 animate-spin rounded-full border-2 border-border border-t-accent" />
+          </div>
+        ) : error ? (
+          <div className="py-20 text-center text-danger">Failed to load runs. Is the backend running?</div>
+        ) : runs && runs.length > 0 ? (
+          <Table>
+            <Thead>
+              <Tr>
+                <Th>Run</Th>
+                <Th>Status</Th>
+                <Th>Started</Th>
+                <Th>By</Th>
+                <Th>Servers</Th>
+              </Tr>
+            </Thead>
+            <Tbody>
+              {runs.map((r: WorkflowRun) => (
+                <Tr key={r.run_id}>
+                  <Td>
+                    <Link
+                      href={`/workflows/${id}/runs/${r.run_id}`}
+                      className="font-mono text-text-primary hover:text-signal"
+                    >
+                      {r.run_id.slice(0, 8)}
+                    </Link>
+                  </Td>
+                  <Td>
+                    <Badge variant={statusVariant[r.status] ?? "neutral"}>{r.status}</Badge>
+                  </Td>
+                  <Td className="text-text-secondary">{new Date(r.started_at).toLocaleString()}</Td>
+                  <Td className="text-text-secondary">{r.triggered_by}</Td>
+                  <Td className="text-text-secondary">{r.server_runs.length}</Td>
+                </Tr>
+              ))}
+            </Tbody>
+          </Table>
+        ) : (
+          <div className="py-16 text-center text-text-secondary">No runs yet.</div>
+        )}
+      </Card>
+    </div>
+  );
+}

web/app/workflows/page.tsx

@@ -82,9 +82,14 @@ export default function WorkflowsPage() {
                     <span className="text-text-secondary">{w.steps.length}</span>
                   </Td>
                   <Td>
-                    <Link href={`/workflows/${w.workflow_id}`}>
-                      <Button variant="ghost" size="sm">Open →</Button>
-                    </Link>
+                    <div className="flex items-center justify-end gap-2">
+                      <Link href={`/workflows/${w.workflow_id}/runs`}>
+                        <Button variant="ghost" size="sm">Runs</Button>
+                      </Link>
+                      <Link href={`/workflows/${w.workflow_id}`}>
+                        <Button variant="ghost" size="sm">Open →</Button>
+                      </Link>
+                    </div>
                   </Td>
                 </Tr>
               ))}

web/components/ui/Modal.tsx

@@ -0,0 +1,45 @@
+"use client";
+
+import { useEffect } from "react";
+
+export function Modal({
+  open,
+  title,
+  onClose,
+  children,
+  wide,
+}: {
+  open: boolean;
+  title: string;
+  onClose: () => void;
+  children: React.ReactNode;
+  wide?: boolean;
+}) {
+  useEffect(() => {
+    if (!open) return;
+    const onKey = (e: KeyboardEvent) => e.key === "Escape" && onClose();
+    window.addEventListener("keydown", onKey);
+    return () => window.removeEventListener("keydown", onKey);
+  }, [open, onClose]);
+
+  if (!open) return null;
+
+  return (
+    <div className="fixed inset-0 z-50 flex items-center justify-center p-4">
+      <div className="absolute inset-0 bg-black/60" onClick={onClose} />
+      <div
+        className={`relative z-10 w-full ${wide ? "max-w-2xl" : "max-w-md"} max-h-[90vh] overflow-auto rounded-xl border border-border bg-surface shadow-2xl`}
+        role="dialog"
+        aria-modal="true"
+      >
+        <div className="flex items-center justify-between border-b border-border px-5 py-3">
+          <h2 className="text-sm font-bold text-text-primary">{title}</h2>
+          <button onClick={onClose} className="text-text-secondary hover:text-text-primary" aria-label="Close">
+            ✕
+          </button>
+        </div>
+        <div className="p-5">{children}</div>
+      </div>
+    </div>
+  );
+}

web/components/ui/index.ts

@@ -2,3 +2,4 @@ export { Button } from "./Button";
 export { Badge } from "./Badge";
 export { Card, CardHeader, CardTitle } from "./Card";
 export { Table, Thead, Tbody, Tr, Th, Td } from "./Table";
+export { Modal } from "./Modal";

web/components/workflows/EditStepModal.tsx

@@ -0,0 +1,110 @@
+"use client";
+
+import { useState } from "react";
+import { useQueryClient } from "@tanstack/react-query";
+import { api, WorkflowStep, InputParam } from "@/lib/api";
+import { Button, Modal } from "@/components/ui";
+
+const inputClass =
+  "w-full rounded-lg border border-border bg-surface-2 px-3 py-2 text-sm text-text-primary placeholder-text-secondary/50 focus:border-signal focus:outline-none focus:ring-1 focus:ring-signal";
+
+export function EditStepModal({ open, step, onClose }: { open: boolean; step: WorkflowStep | null; onClose: () => void }) {
+  const qc = useQueryClient();
+  const [name, setName] = useState(step?.name ?? "");
+  const [interpreter, setInterpreter] = useState<"bash" | "powershell">(step?.interpreter ?? "bash");
+  const [script, setScript] = useState(step?.script ?? "");
+  const [outputs, setOutputs] = useState<string[]>(step?.declared_outputs ?? []);
+  const [inputs, setInputs] = useState<InputParam[]>(step?.declared_inputs ?? []);
+  const [newOut, setNewOut] = useState("");
+  const [busy, setBusy] = useState(false);
+  const [error, setError] = useState<string | null>(null);
+
+  // NOTE: because state is seeded from props, render the modal conditionally
+  // (parent mounts it only when opening) OR key it by step_id so it re-seeds.
+
+  const save = async () => {
+    setBusy(true); setError(null);
+    try {
+      const payload: Partial<WorkflowStep> = {
+        name: name.trim(), description: step?.description ?? "", interpreter, script,
+        declared_outputs: outputs, declared_inputs: inputs.filter((i) => i.name.trim() !== ""),
+        secret_refs: step?.secret_refs ?? [],
+      };
+      if (step) await api.updateStep(step.step_id, payload);
+      else await api.createStep(payload);
+      qc.invalidateQueries({ queryKey: ["steps"] });
+      onClose();
+    } catch (e) { setError((e as Error).message); } finally { setBusy(false); }
+  };
+
+  const del = async () => {
+    if (!step || !window.confirm("Delete this step? It will be removed from every workflow that uses it.")) return;
+    setBusy(true); setError(null);
+    try {
+      await api.deleteStep(step.step_id);
+      qc.invalidateQueries({ queryKey: ["steps"] });
+      qc.invalidateQueries({ queryKey: ["workflow"] });
+      onClose();
+    } catch (e) { setError((e as Error).message); } finally { setBusy(false); }
+  };
+
+  return (
+    <Modal open={open} onClose={onClose} title={step ? "Edit base step" : "New step"} wide>
+      <div className="space-y-4">
+        {error && <div className="rounded border border-danger/30 bg-danger/10 px-3 py-2 text-sm text-danger">{error}</div>}
+        <p className="text-xs text-text-secondary">Reusable steps are shared across all workflows. Editing here changes it everywhere.</p>
+        <div>
+          <label className="mb-1 block text-xs uppercase text-text-secondary">Name</label>
+          <input className={inputClass} value={name} onChange={(e) => setName(e.target.value)} />
+        </div>
+        <div>
+          <label className="mb-1 block text-xs uppercase text-text-secondary">Interpreter</label>
+          <select className={inputClass} value={interpreter} onChange={(e) => setInterpreter(e.target.value as "bash" | "powershell")}>
+            <option value="bash">bash</option>
+            <option value="powershell">powershell</option>
+          </select>
+        </div>
+        <div>
+          <label className="mb-1 block text-xs uppercase text-text-secondary">Script</label>
+          <textarea className={`${inputClass} h-40 font-mono text-xs`} value={script} onChange={(e) => setScript(e.target.value)} />
+          <p className="mt-1 text-xs text-text-secondary">Write <code className="text-signal">KEY=value</code> to <code className="text-signal">$WORKFLOW_ENV</code> to expose it to later steps.</p>
+        </div>
+        <div>
+          <label className="mb-1 block text-xs uppercase text-text-secondary">Outputs</label>
+          <div className="mb-2 flex flex-wrap gap-1">
+            {outputs.map((o) => (
+              <span key={o} className="flex items-center gap-1 rounded bg-signal px-2 py-0.5 font-mono text-[11px] text-signal-ink">
+                {o}<button onClick={() => setOutputs(outputs.filter((x) => x !== o))}>✕</button>
+              </span>
+            ))}
+          </div>
+          <div className="flex gap-2">
+            <input className={inputClass} placeholder="OUTPUT_NAME" value={newOut} onChange={(e) => setNewOut(e.target.value)} />
+            <Button variant="ghost" size="sm" onClick={() => { if (newOut.trim()) { setOutputs([...outputs, newOut.trim()]); setNewOut(""); } }}>Add</Button>
+          </div>
+        </div>
+        <div>
+          <label className="mb-1 block text-xs uppercase text-text-secondary">Inputs</label>
+          <div className="space-y-2">
+            {inputs.map((inp, i) => (
+              <div key={i} className="flex gap-2">
+                <input className={inputClass} placeholder="name" value={inp.name} onChange={(e) => setInputs(inputs.map((x, j) => j === i ? { ...x, name: e.target.value } : x))} />
+                <input className={inputClass} placeholder="default" value={inp.default} onChange={(e) => setInputs(inputs.map((x, j) => j === i ? { ...x, default: e.target.value } : x))} />
+                <input className={inputClass} placeholder="description" value={inp.description} onChange={(e) => setInputs(inputs.map((x, j) => j === i ? { ...x, description: e.target.value } : x))} />
+                <Button variant="ghost" size="sm" onClick={() => setInputs(inputs.filter((_, j) => j !== i))}>✕</Button>
+              </div>
+            ))}
+          </div>
+          <Button variant="ghost" size="sm" className="mt-2" onClick={() => setInputs([...inputs, { name: "", default: "", description: "" }])}>Add input</Button>
+        </div>
+        <div className="flex items-center justify-between pt-2">
+          {step ? <Button variant="danger" onClick={del} loading={busy}>Delete step</Button> : <span />}
+          <div className="flex gap-2">
+            <Button variant="ghost" onClick={onClose}>Cancel</Button>
+            <Button variant="primary" onClick={save} loading={busy} disabled={!name.trim()}>Save</Button>
+          </div>
+        </div>
+      </div>
+    </Modal>
+  );
+}

web/components/workflows/EditWorkflowModal.tsx

@@ -0,0 +1,77 @@
+"use client";
+
+import { useState, useEffect } from "react";
+import { useRouter } from "next/navigation";
+import { useQuery } from "@tanstack/react-query";
+import { api, Workflow } from "@/lib/api";
+import { Button, Modal } from "@/components/ui";
+
+const inputClass =
+  "w-full rounded-lg border border-border bg-surface-2 px-3 py-2 text-sm text-text-primary focus:border-signal focus:outline-none focus:ring-1 focus:ring-signal";
+
+export function EditWorkflowModal({ open, workflow, onSaved, onClose }: { open: boolean; workflow: Workflow; onSaved: (w: Workflow) => void; onClose: () => void }) {
+  const router = useRouter();
+  const [name, setName] = useState(workflow.name);
+  const [targets, setTargets] = useState<string[]>(workflow.target_server_ids);
+  const [busy, setBusy] = useState(false);
+  const [error, setError] = useState<string | null>(null);
+  const { data: servers } = useQuery({ queryKey: ["servers"], queryFn: api.listServers });
+
+  useEffect(() => {
+    if (open) {
+      setName(workflow.name);
+      setTargets(workflow.target_server_ids);
+    }
+  }, [open, workflow]);
+
+  const toggle = (id: string) => setTargets((t) => (t.includes(id) ? t.filter((x) => x !== id) : [...t, id]));
+
+  const save = async () => {
+    setBusy(true); setError(null);
+    try {
+      const updated = await api.updateWorkflow(workflow.workflow_id, { ...workflow, name, target_server_ids: targets });
+      onSaved(updated); onClose();
+    } catch (e) { setError((e as Error).message); } finally { setBusy(false); }
+  };
+
+  const del = async () => {
+    if (!window.confirm("Delete this workflow? This cannot be undone.")) return;
+    setBusy(true); setError(null);
+    try { await api.deleteWorkflow(workflow.workflow_id); router.push("/workflows"); }
+    catch (e) { setError((e as Error).message); setBusy(false); }
+  };
+
+  return (
+    <Modal open={open} onClose={onClose} title="Edit workflow">
+      <div className="space-y-4">
+        {error && <div className="rounded border border-danger/30 bg-danger/10 px-3 py-2 text-sm text-danger">{error}</div>}
+        <div>
+          <label className="mb-1 block text-xs uppercase text-text-secondary">Name</label>
+          <input className={inputClass} value={name} onChange={(e) => setName(e.target.value)} />
+        </div>
+        <div>
+          <label className="mb-1 block text-xs uppercase text-text-secondary">Target servers</label>
+          <div className="flex flex-wrap gap-2">
+            {servers?.map((s) => {
+              const on = targets.includes(s.server_id);
+              return (
+                <label key={s.server_id} className={`flex cursor-pointer items-center gap-2 rounded-lg border px-2 py-1 text-sm ${on ? "border-signal bg-signal/10 text-text-primary" : "border-border text-text-secondary"}`}>
+                  <input type="checkbox" className="accent-signal" checked={on} onChange={() => toggle(s.server_id)} />
+                  {s.hostname}
+                </label>
+              );
+            })}
+            {servers && servers.length === 0 && <p className="text-xs text-text-secondary">No servers registered.</p>}
+          </div>
+        </div>
+        <div className="flex items-center justify-between pt-2">
+          <Button variant="danger" onClick={del} loading={busy}>Delete workflow</Button>
+          <div className="flex gap-2">
+            <Button variant="ghost" onClick={onClose}>Cancel</Button>
+            <Button variant="primary" onClick={save} loading={busy} disabled={!name.trim()}>Save</Button>
+          </div>
+        </div>
+      </div>
+    </Modal>
+  );
+}

web/lib/api.ts

@@ -95,6 +95,7 @@ export interface Settings {
   alerts: AlertSettings;
   email: EmailSettings;
   secrets: SecretsSettings;
+  workflow_log_retention_days?: number | null;
 }
 
 export interface SecretGroupSummary {
@@ -132,6 +133,12 @@ export interface ServerWithKeys extends Server {
   keys: (Assignment & { key: Key })[];
 }
 
+export interface InputParam {
+  name: string;
+  default: string;
+  description: string;
+}
+
 export interface WorkflowStep {
   step_id: string;
   name: string;
@@ -139,6 +146,7 @@ export interface WorkflowStep {
   interpreter: "bash" | "powershell";
   script: string;
   declared_outputs: string[];
+  declared_inputs: InputParam[];
   secret_refs: string[];
 }
 
@@ -148,6 +156,7 @@ export interface WorkflowStepRef {
   on_failure: "stop" | "continue" | "retry";
   max_retries: number;
   overrides?: { script?: string; secret_refs?: string[] };
+  inputs?: Record<string, string>;
 }
 
 export interface Workflow {
@@ -163,8 +172,7 @@ export interface StepRun {
   status: string;
   attempts: number;
   exit_code: number;
-  stdout: string;
-  stderr: string;
+  log_offset: number;
   output_env: Record<string, string>;
   started_at?: string;
   finished_at?: string;
@@ -282,7 +290,11 @@ export const api = {
     return request<Settings>("/settings");
   },
 
-  saveSettings(settings: { alerts: AlertSettings; email: EmailSettings }): Promise<{ saved: boolean }> {
+  saveSettings(settings: {
+    alerts: AlertSettings;
+    email: EmailSettings;
+    workflow_log_retention_days?: number | null;
+  }): Promise<{ saved: boolean }> {
     return request<{ saved: boolean }>("/settings", {
       method: "PUT",
       body: JSON.stringify(settings),
@@ -452,4 +464,16 @@ export const api = {
   cancelRun(runId: string): Promise<void> {
     return request<void>(`/runs/${runId}/cancel`, { method: "POST" });
   },
+
+  async getServerRunLog(runId: string, serverId: string): Promise<string> {
+    const res = await fetch(`/api/runs/${runId}/servers/${serverId}/logs`, {
+      credentials: "include",
+    });
+    if (!res.ok) throw new Error("no logs");
+    return res.text();
+  },
+
+  serverRunLogStreamUrl(runId: string, serverId: string): string {
+    return `/api/runs/${runId}/servers/${serverId}/logs/stream`;
+  },
 };

web/tailwind.config.ts

@@ -21,6 +21,10 @@ const config: Config = {
         warning: "#f59e0b",
         danger: "#ef4444",
         "danger-hover": "#dc2626",
+        bash: "#3fb950",
+        pwsh: "#5b9bff",
+        signal: "#f5a524",
+        "signal-ink": "#241800",
       },
     },
   },