Skip to main content
skill CX Codex

workspace · sethgammon — Claude Code skill: multi-repo campaign coordinator

workspace coordinates changes across multiple repositories in a single task. Breaks work into waves, runs campaigns per repo, verifies contracts between them.

MIT ★ 937 updated 2026-10-29

Install into Claude Code · Plugin marketplace All methods and platforms ↓
/plugin marketplace add SethGammon/Citadel
/plugin install citadel@citadel-local
Open on GitHub

What workspace does

Breaks a task into repo-scoped campaigns and runs them in waves: dependencies first, consumers second. For each repo, creates a branch, spawns /archon, /fleet or /marshal with the direction, collects results and relays context to the next wave. Before spawning consumers, verifies that producers delivered the promised artefacts — types in a package, API routes, environment variables.

If one repo fails, others continue: no automatic rollback, user decides. All repos must be locally accessible and be git repositories. For monorepos, each package counts as a separate work unit.

What it helps with

  • Add a new database or service that touches frontend, backend and infrastructure
  • Split a monolith into services with coordinated contracts
  • Roll out an API contract change through all dependent repositories
  • Synchronise type and route updates across shared-types, backend and frontend
  • Deploy a new microservice with coordinated CI/CD and config changes

Example prompts for the agent

  • Add Redis to backend-repo and set up the client in frontend-repo using shared types from shared-types
  • Split the monolith in ./app into three services: auth-service, api-service, worker-service with shared types
  • Update the /users endpoint in backend and the corresponding hook in frontend, types in types-package first

How to install workspace

1Plugin marketplace

/plugin marketplace add SethGammon/Citadel
/plugin install citadel@citadel-local

This installs the whole citadel plugin with all its skills

2Skills CLI (Vercel)

npx skills add SethGammon/Citadel --skill workspace -a claude-code

3Manually: put the files in place

~/.claude/skills/workspace/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/workspace/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about workspace

How to install workspace in Claude Code?

Run /plugin marketplace add SethGammon/Citadel, then /plugin install citadel@citadel-local. Or via CLI: npx skills add SethGammon/Citadel --skill workspace -a claude-code.

What does workspace do?

Coordinates changes across multiple repositories: breaks a task into campaigns, runs them in waves respecting dependencies, verifies contracts between repos and relays context between waves.

What does workspace require to work?

All repos must be locally accessible and be git repositories. Node.js to run scripts from .citadel/scripts/. If a repo has an active campaign, workspace reports the conflict.

Source text

SKILL.md · MIT
---
name: workspace
license: MIT
description: >-
  Multi-repo campaign coordinator. Same lifecycle as fleet -- scope claims,
  discovery relay, wave-based execution -- but the unit of work is a repo,
  not a file. Coordinates campaigns across repositories with shared context.
user-invocable: true
auto-trigger: false
trigger_keywords:
  - workspace
  - multi-repo
  - cross-repo
  - across repos
  - multiple repos
  - coordinate repos
  - add redis and snowflake
  - split into repos
last-updated: 2026-03-29
effort: high
---

# /workspace -- Multi-Repo Campaign Coordinator

## When to Use

- Adding infrastructure that spans repos (new database, shared service, API contract)
- Coordinating changes across a frontend repo, backend repo, and infra repo
- Breaking a monolith into services (each service becomes a repo-scoped campaign)
- Any task where changes in repo A depend on or inform changes in repo B

**Do not use when:**
- All work is in one repo (use `/fleet` or `/archon`)
- The repos are truly independent with no shared contracts (just run separate campaigns)

## Protocol

### Step 1: ORIENT

1. Check for existing workspace session: `.planning/workspace/session-{slug}.md` — resume if `status: active` or `needs-continue`
2. If starting fresh: identify repos, verify each path is a git repo, read each repo's `CLAUDE.md`, check `.planning/campaigns/` for active campaigns (avoid collisions)
3. **Load prior session context and start watcher**:
   ```bash
   node .citadel/scripts/momentum-watch-start.cjs
   node .citadel/scripts/momentum-read.cjs
   ```
   Skip momentum injection if output is empty.

### Step 2: DECOMPOSE

Break the direction into repo-scoped work items (one campaign per repo per wave). Table format: `# | Repo | Campaign Direction | Scope | Deps | Wave`.

**Rules:** items with no deps → Wave 1; dependents → Wave 2+. Max 3 repo-campaigns per wave. Scope format: `{repo}:{path}`.

For each inter-wave dependency, specify the cross-repo contract: what the producer will produce, what the consumer expects, and where the contract lives (shared types package, OpenAPI spec, env var).

### Step 3: WORKSPACE SESSION FILE

Create `.planning/workspace/session-{slug}.md` with frontmatter: `version`, `id`, `status: active`, `started`, `completed_at: null`, `direction`, `repos` (path + name + branch per repo), `wave_count`, `current_wave: 1`, `campaigns_total`, `campaigns_complete: 0`.

Body sections: Direction, Repos table (name/path/branch/status), Work Queue (table from Step 2), Cross-Repo Contracts (producer/consumer/contract/location), Wave Execution Log (per wave: status, campaigns, started, completed), Shared Context (discovery relay accumulation).

### Step 4: WAVE EXECUTION

For each wave:

#### 4a. Pre-flight
- Verify all dependency campaigns from prior waves completed successfully
- Check cross-repo contracts: did producer repos create the expected outputs?
- If a dependency failed: park the dependent campaign, flag for user decision

#### 4b. Spawn campaigns
For each repo-campaign in this wave:

1. Create branch: `git checkout -b workspace/{slug}/{repo-name}`
2. Spawn agent with direction: `/archon` for complex (3+ phases), `/fleet` if parallelizable, `/marshal` or direct skill for simple (1-2 steps)
3. Inject context: discovery briefs from prior waves, prior session context (re-read `momentum.json` via `node .citadel/scripts/momentum-read.cjs`, inject as `=== PRIOR SESSION CONTEXT ===`, skip if empty), cross-repo contracts, relevant `CLAUDE.md` sections from other repos

#### 4c. Collect results
Extract HANDOFF blocks, compress into cross-repo discovery brief, write persistent discovery records:
  ```bash
  node .citadel/scripts/discovery-write.cjs \
    --session {session-slug} --agent {repo-name}-{campaign-type} \
    --wave {wave-number} --status {success|partial|failed} \
    --scope "{repo-name}:{scope-path}" --handoff "{json-array}" \
    --decisions "{json-array}" --files "{json-array}" --failures "{json-array}"
  ```

#### 4d. Discovery relay
Write `workspace/briefs/wave{N}-{repo-name}.md` for each completed campaign.
Also write `workspace/briefs/wave{N}-cross-repo.md` summarizing:
- New API endpoints or types created
- Config changes that affect other repos
- Contract fulfillment status (did the producer deliver what was promised?)

#### 4e. Contract verification
For each cross-repo contract: verify the producer created the expected output (file + export for types, route existence for endpoints). If verification fails, flag and do not proceed with consumers.

#### 4f. Update session
Mark completed campaigns, update wave status, write discovery relay, advance `current_wave`.

### Step 5: COMPLETION

1. Run typecheck/build for each repo in isolation (build shared types package first if present)
2. Set session `status: completed`, `completed_at: {ISO timestamp}`
3. Run `node .citadel/scripts/momentum-synthesize.cjs`
4. List all branches created across repos with suggested merge order based on dependency graph
5. Output HANDOFF

## Fringe Cases

- **Repo not a git repo:** Skip it. Report which repos were skipped and why.
- **Repo has uncommitted changes:** Stash before branching. Record stash ref in session file.
  Pop on completion or failure.
- **Active campaign in target repo:** Do not start a second campaign. Report the conflict
  and ask the user whether to wait, park the existing campaign, or merge scopes.
- **`.planning/workspace/` does not exist:** Create it (and `workspace/briefs/`).
- **Cross-repo contract broken:** Park all downstream campaigns. Report which contract
  failed, which producer was responsible, and what the consumer expected. Do not
  attempt to fix the producer -- surface the issue for the user or re-run the producer campaign.
- **One repo fails, others succeed:** Mark the failed repo-campaign. Do not roll back
  successful repos. The user decides whether to fix-and-continue or abandon.
- **Repos on different machines or remotes:** Not supported. All repos must be locally
  accessible. If a repo is remote-only, the user must clone it first.
- **Monorepo with multiple packages:** Treat each package as a "repo" for scoping purposes.
  Use `{monorepo}:{package-path}` as the scope identifier.

## Contextual Gates

**Disclosure:** "Running multi-repo campaign across [repos]. Changes committed to each repo independently."
**Reversibility:** red — coordinates changes across multiple repositories; cross-repo commits are hard to revert in bulk.
**Trust gates:**
- Familiar (5+ sessions): coordinates multi-repo campaigns autonomously; novices should use /marshal per repo.

## Quality Gates

- [ ] All repos verified as accessible git repositories before starting
- [ ] Work queue has no scope overlaps within the same repo
- [ ] Cross-repo contracts specified for every inter-wave dependency
- [ ] Discovery relay written after each wave
- [ ] Contract verification run before spawning consumer campaigns
- [ ] Each repo's typecheck/build passes independently after completion
- [ ] Session file updated after every wave (not just at the end)
- [ ] No campaigns left in `active` state on completion

## Exit Protocol

```
---HANDOFF---
- Workspace: {slug} -- {direction summary}
- Repos: {N} repos, {M} campaigns across {W} waves
- Results: {completed}/{total} campaigns succeeded
- Branches: {list branches ready for review}
- Merge order: {suggested order based on dependency graph}
- Unresolved: {any failed campaigns or broken contracts}
---
```

© SethGammon · MIT · SethGammon/Citadel

Similar skills

All skills for Agent setup: memory, context, orchestration →

workspace by other authors comes first.

workspace alinaqi Discovers workspace topology across multi-repo and monorepo setups, tracking API contracts and shared types for coding agents. CX Codex +1 198installs multi-repo-swarm ruvnet multi-repo-swarm — a subagent for Claude Code that orchestrates multiple agents across your organization's repositories. Runs synchronized tasks, manages dependencies, and links pull requests between projects. CX Codex ★ 850repo stars habitat-usage lynx-community habitat-usage — a skill for Claude Code, Codex, Cursor and other platforms that configures Habitat to sync multi-repo dependencies and external assets via .habitat/DEPS and hab sync. CX Codex +1 253installs nx-workspace nrwl A skill for Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, and OpenCode that gives AI agents the commands to navigate an Nx workspace. CX Codex +1 100,600installs link-workspace-packages nrwl link-workspace-packages — a skill that makes the agent add workspace dependencies between monorepo packages. Use it when sibling imports fail to resolve and packages need linking through the package manager. CX Codex +1 3,200installs multi-repo-git-ops bmad-labs The multi-repo-git-ops Claude Code skill coordinates Git work across a parent repository and its submodules. It handles branch creation, commits, pushes, and submodule pointer updates. CX Codex +1 104installs pnpm-workspace oakoss pnpm-workspace teaches coding agents to configure pnpm monorepos and version catalogs. Useful when managing multi-package repositories. CX Codex +1 195installs

From the same set55

The whole set →
postmortem The postmortem skill reviews completed campaigns or incidents using project files, telemetry, and Git history. It shows what failed and which safeguards caught it. CX Codex +1 ★ 937repo stars map map generates a JSON index of a repository's files, exports, imports and dependency graph so an agent can locate relevant code without extra Glob or Grep calls. CX Codex +1 ★ 937repo stars phase-validator The phase-validator subagent for Claude Code compares phase handoff reports with exit conditions and returns a pass or fail verdict. It checks automated development pipelines. CX Codex ★ 937repo stars fleet fleet splits a big job into campaigns and runs them in waves of 2–3 agents, handing each wave's findings to the next. Fits when the tasks don't touch the same files. CX Codex ★ 937repo stars create-skill create-skill interviews you about a recurring workflow, compiles it into a structured SKILL.md file, and tests it against real project files. CX Codex +1 ★ 937repo stars arbiter arbiter is a Claude Code subagent that reruns project checks and gives a final verdict on changes. It steps in after a worker runs out of retries or when a run needs a holistic completion review. CX Codex ★ 937repo stars infra-audit infra-audit maps a project's databases, caches and queues from docker-compose, env files and ORM configs, then flags infrastructure layers that look missing. CX Codex +1 ★ 937repo stars refactor Multi-file refactoring with automated Git rollback, typechecking, and test baseline verification across edits. CX Codex +1 ★ 937repo stars

Project sponsors

Companies that keep this analytics open