puppet-master
/puppet-master
Deleguje pracę innym agentom programistycznym (Claude Code, Codex, Cursor, OpenCode) przez narzędzia MCP Puppet Master: druga opinia, równoległe lub ryzykowne zmiany w izolowanym git worktree, kontrola działających agentów.
- Autor
- potato-dev-inc ↗
- Repozytorium
- puppet-master ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 28.10.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 1
- Skrypty
- 0
- Komentarze HTML
- 1
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 28.10.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
- SKILL.md zawiera komentarze HTML (1). Nie widać ich w podglądzie Markdown, a model je czyta.
Co robi i kiedy
Deleguje pracę innym agentom programistycznym (Claude Code, Codex, Cursor, OpenCode) przez narzędzia MCP Puppet Master: druga opinia, równoległe lub ryzykowne zmiany w izolowanym git worktree, kontrola działających agentów.
Oryginalny opis (EN)
Delegate work to other AI coding agents (Claude Code, Codex, Cursor, OpenCode) through the Puppet Master MCP tools. Use it for a second opinion or review from a different agent, a quick question to another model, a parallel or risky change in an isolated git worktree, or to check on agents that are already running. Needs the Puppet Master desktop app to be running.
Jak działa
- When to use it
- The five tools
- Do not trust "done"
- Isolate risky or parallel edits
- Ask the human
- Etiquette
Podgląd SKILL.md
---
name: puppet-master
description: Delegate work to other AI coding agents (Claude Code, Codex, Cursor, OpenCode) through the Puppet Master MCP tools. Use it for a second opinion or review from a different agent, a quick question to another model, a parallel or risky change in an isolated git worktree, or to check on agents that are already running. Needs the Puppet Master desktop app to be running.
---
<!-- puppet-master-pack: 1 -->
# Delegating with Puppet Master
Puppet Master (PM) lets you hand work to other coding agents and get back a result plus a receipt that says what was and was not verified. Your host may show its tools with a prefix (for example `mcp__puppet-master__run_agent`). If they are missing, the PM app is not running: tell the user, do not guess.
## When to use it
- A **second opinion** from a different agent before you call work done.
- A **tiny question** you want answered by another model.
- A **parallel** or **risky** change that should not touch the user's files yet.
- **Checking on** agents that are already running.
Do not delegate what you can do directly in a few steps, and never put secrets in a task.
## The five tools
1. **`quick_ask(task)`** for a tiny question. Read `answer`. `receipt.not_verified` lists what was not checked. If it returns `status: "running"`, call `wait_agents` with the `handle`.
2. **`run_agent(agent_type, project_path, task)`** for real work. `agent_type` is `claude`, `codex`, `cursor_agent` or `opencode_native`. It opens a visible pane the user can watch (`headless: true` hides it), runs the task and returns `status`, `result`, `receipt` and a `handle`. Pick a different agent than yourself for a review.
3. **`wait_agents(handles)`** collects results. Each call waits at most 120 s, so loop while `status` is `running`. A turn that still runs carries a `watch_command`: run it in your shell tool's background mode.
4. **`followup_task(handle, task)`** is the next turn with the same agent and the same conversation. `send_message(handle, message)` steers the turn that is running now.
5. **`list_agents()`** shows agents you can reuse: pass the `worker_id` to `run_agent` to adopt one.
## Do not trust "done"
Read `receipt.outcome`. Only `passed_checks` means the result was proven. `completed_unchecked` means the agent finished and nothing verified it.
Pass `checks` to `run_agent` so success is measured, for example:
```json
{ "checks": [
{ "type": "command_exits", "command": ["npm", "test"] },
{ "type": "file_exists", "path": "src/new-file.ts" }
] }
```
Other check types: `file_contains` (with `text`), `file_absent`, `result_contains`. A failed check ends the run as `ACCEPTANCE_FAILED`. When you report back, say what the receipt did not verify.
## Isolate risky or parallel edits
Pass `isolation: "worktree"` to `run_agent`. The agent works on its own branch `pm/<handle>` under `.puppet-master/worktrees`, so parallel agents never touch the same files. The receipt gains `worktree` with the change counts. Nothing is merged for you:
- `review_worktree(handle)` shows the diff.
- `review_worktree(handle, action: "merge")` merges it. `action: "discard"` deletes the work.
Merge or discard only when the user has said so.
## Ask the human
- `ask_user(question, choices?)` when only the person can decide. It shows in the PM Inbox and on a paired phone, and blocks until answered. Do not ask what you can find out yourself.
- If a run ends `needs_input`, it carries a `prompt`. Show it to the user with `ask_user`, then pass their choice to `answer_prompt(handle, prompt_id, choice)`. Never choose approvals for them.
## Etiquette
- Use `read_only: true` for reviews and questions.
- `interrupt_agent(handle)` stops a turn and keeps the handle.
- Do not close the user's own panes. Panes a run created close themselves; pass `keep_pane: true` to keep one.
- Reuse the same `idempotency_key` when you retry after a timeout, so no duplicate turn starts.
Wersje
W bazie nie ma innych skilli o nazwie puppet-master.
Identyczna treść występuje jeszcze w 3 innych folderach na GitHubie. Tu pokazujemy jedną, najlepiej ocenioną kopię.
Migawka bazy z 13.11.2026 · wersja paczki 7401d6c49c94 (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 100potato-workspace
Deleguje zadania innym agentom kodującym przez narzędzia MCP Potato Workspace: drugą opinię, równoległe zmiany w worktree i podgląd agentów. - 100opencode-subagents
Deleguje zadania do opencode jako zewnętrznego podagenta na modelach innych niż Claude i OpenAI, także równolegle w worktree. - 100nuncius
Przekazuje zadania kodowania do Codexa, Cursora lub innego modelu Claude przez Nuncius, na własnych subskrypcjach użytkownika. - 100opencode-delegate
Zleca zadanie kodowe lub pytanie modelowi DeepSeek przez opencode, a potem sprawdza wynik jako tani drugi model. - 100orca
Obsługuje aplikację Orca, która uruchamia kilku agentów kodujących, takich jak Claude Code i Codex, równolegle w osobnych git worktree, na przykład do porównania wyników na tym samym zadaniu. - 100superagent
Zarządza pracą kodową jak router: przydziela zadania agentom Claude Code, Codex, OpenCode i Cursor w osobnych worktree, równolegle.