Project Workbench Skill

A skill for managing project registries and work sessions across multiple workspaces, with support for parallel sessions and session resume.

Configuration

The workbench registry is stored at ~/.claude/skills/project-workbench/workbench.json.

Read it before any operation:

cat ~/.claude/skills/project-workbench/workbench.json

User Commands

How to Use This Skill

Step 1: Read Registry

Always start by reading the workbench registry:

cat ~/.claude/skills/project-workbench/workbench.json

If the file does not exist, create it by running:

bash ~/.claude/skills/project-workbench/scripts/init-workbench.sh

Step 2: Parse Command

Parse the user's /workbench invocation to determine the sub-command and arguments.

Step 3: Execute Operation

Dashboard (/workbench)

1. •Read workbench.json
2. •For each project with status != "archived":
   • •Count active/paused sessions
   • •List workspace_dirs
3. •Print a formatted dashboard showing projects and their active sessions

Format:

  PROJECT-NAME (status)  ·  N workspaces  ·  N active sessions
  |- ses_..HHMM  [tool]  dir-label  "intent text"
  \- ses_..HHMM  [tool]  dir-label  "intent text"

Add Project (/workbench add <name>)

1. •Ask user for: description, workspace directories (label:path pairs), repo URL (optional)
2. •Generate slug from name (lowercase, hyphenated)
3. •Get timestamp: date -u +%Y-%m-%dT%H:%M:%SZ
4. •Create project entry in workbench.json with status "active"
5. •Write updated workbench.json

Add Workspace Dir (/workbench add-dir <project> <label> <path>)

1. •Validate project exists
2. •Add label: path to project's workspace_dirs
3. •Write updated workbench.json

Start Session (/workbench start <project>)

Options: --dir <label>, --tool <tool>, --intent "<text>", --branch <branch>

1. •Validate project exists and is active
2. •If --intent not provided, ask user what they plan to work on
3. •Generate session ID using date: ses_$(date +%Y%m%d_%H%M%S)
4. •Default --tool to claude-code if not specified
5. •Create session object, append to project.sessions[]
6. •Write updated workbench.json
7. •Print session context

Resume Session (/workbench resume [session_id])

1. •If no session_id, find most recent paused session (by updated timestamp, across all projects)
2. •Load session data
3. •Set status -> "active", update timestamp using date -u +%Y-%m-%dT%H:%M:%SZ
4. •Print full context block:
   • •Project name, workspace dir path, intent, last summary
   • •Branch, open files, next steps (if set)
5. •Write updated workbench.json
6. •This context block gives Claude Code / Codex enough info to continue

Switch Session (/workbench switch <session_id>)

1. •Find current active session (if any) and auto-pause it
   • •If no summary set, note "auto-paused, no summary captured"
2. •Find target session, set status -> "active"
3. •Print resume context for target session
4. •Write updated workbench.json

Pause / Done (/workbench pause|done)

1. •Find current active session
2. •If --summary provided, use it; otherwise ask user for 1-3 sentence summary
3. •Set status -> "paused" (or "done")
4. •Update timestamp using date -u +%Y-%m-%dT%H:%M:%SZ
5. •Optionally ask for next_steps
6. •Write updated workbench.json

List Sessions (/workbench list)

Options: --project <slug>, --status <status>

1. •Read workbench.json
2. •Filter sessions by project and/or status
3. •Print table: ID | Project | Status | Tool | Dir | Intent

List Projects (/workbench list projects)

1. •Read workbench.json
2. •Print table: Slug | Name | Status | Workspaces | Active Sessions

Archive (/workbench archive <target>)

1. •Find project or session by slug/id
2. •Set status -> "archived"
3. •Write updated workbench.json

Sync (/workbench sync)

1. •Read config to check if a sync remote is configured
2. •cd to the directory containing workbench.json
3. •git add workbench.json && git commit -m "workbench sync $(date +%Y-%m-%d %H:%M)" && git push

Step 4: Write Back

After every mutation, write the updated workbench.json back to disk using the Write tool.

IMPORTANT: Use date command for all timestamps. Never guess or make up times.

Summary Guidelines

Summaries must be:

• •1-3 sentences max -- high-level, not granular
• •Outcome-oriented -- what was accomplished or what state things are in
• •Actionable -- mention what's next if relevant

Good: "Implemented OAuth2 token refresh. Login flow works end-to-end. Logout endpoint still needs wiring up." Bad: "Edited oauth2.ts lines 45-120, added refreshToken function, modified token-store.ts..."

Session Object Schema

{
  "id": "ses_20260209_143000",               // deterministic: ses_YYYYMMDD_HHMMSS
  "project": "openclaw",
  "status": "active",                        // active | paused | done | archived
  "tool": "claude-code",                     // claude-code | codex | cursor | manual
  "workspace_dir": "server",                 // key from project.workspace_dirs
  "created": "2026-02-09T14:30:00Z",
  "updated": "2026-02-09T16:45:00Z",
  "intent": "Implement OAuth2 flow for SSO",
  "summary": "",                             // filled on pause/done
  "branch": "",                              // optional: git branch
  "open_files": [],                          // optional: breadcrumbs for resume
  "blockers": [],                            // optional
  "next_steps": ""                           // optional: filled on pause/done
}

Integration with memo Skill

• •project-workbench handles what project/session am I in (orchestration)
• •memo handles what did I learn/decide/do (documentation)
• •They share the same project slug namespace
• •A session's intent and summary are NOT duplicated in memo -- they serve different purposes