claude-delegate
/claude-delegate
Kieruje zadanie do natywnego subagenta Claude na dopasowanym modelu i wybiera między nim a zewnętrznymi pracownikami, takimi jak Codex lub Gemini.
- Autor
- joezhuo2 ↗
- Repozytorium
- delegate-skills ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 11.11.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 1
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 11.11.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Kieruje zadanie do natywnego subagenta Claude na dopasowanym modelu i wybiera między nim a zewnętrznymi pracownikami, takimi jak Codex lub Gemini.
Oryginalny opis (EN)
Route a task to a native Claude subagent on the model that fits it (Haiku, Sonnet, Opus or Fable), and decide between a Claude subagent and the external delegate workers (Codex, Kiro, Gemini, Qwen, Copilot). Use whenever you are about to hand off work or spawn a subagent. Follow the delegate-skills level (low, medium, ultra) given in session context.
Jak działa
- Delegation level
- Choose the worker
- Choose the Claude model
- Prompt template
- After the subagent returns (mandatory)
Podgląd SKILL.md
--- name: claude-delegate description: Route a task to a native Claude subagent on the model that fits it (Haiku, Sonnet, Opus or Fable), and decide between a Claude subagent and the external delegate workers (Codex, Kiro, Gemini, Qwen, Copilot). Use whenever you are about to hand off work or spawn a subagent. Follow the delegate-skills level (low, medium, ultra) given in session context. --- # claude-delegate Spawn a Claude subagent with the Agent tool on the cheapest model that can do the task well. Claude stays orchestrator and reviewer, and decides per task which worker gets it. ## Delegation level Set with `/delegate low|medium|ultra` (default `medium`); a SessionStart hook prints the saved level and the enabled workers into each session. - `low`: delegate only when the user explicitly asks. - `medium`: delegate when the task fits the rules below. - `ultra`: prefer delegating over doing the work yourself. Delegate anything a worker can do; keep orchestration, review and work no worker can do. If `claude` is not in the enabled list, this part is disabled: do not use this skill's routing. ## Choose the worker You are the orchestrator. For each task, pick one: | Task | Worker | | --- | --- | | Needs this conversation's context, back-and-forth, or tools only this session has | Claude subagent (or yourself) | | Self-contained, bounded, verifiable with one command, no conversation context needed | An enabled external worker (`codex-delegate`, `kiro-delegate`, `gemini-delegate`, `qwen-delegate`, `copilot-delegate`), or a Claude subagent if none is enabled or its gate blocks | | Independent second opinion or review of Claude's own work | An external worker | | Several independent tasks at once | Spread them across workers; never two writers on the same files | | An external worker failed or returned a big miss | Claude subagent, one model tier above what the task would normally get | ## Choose the Claude model Pass the Agent tool's `model` parameter. Start at the lowest tier that fits; go up one tier after a failure, not before. | Model | `model` | Give it | | --- | --- | --- | | Haiku | `haiku` | Boilerplate, docs, version bumps, changelog, formatting, renames, log parsing, test scaffolding, single-file mechanical edits, simple lookups | | Sonnet | `sonnet` | Medium features with a clear spec, multi-file bug fixes, localized refactors, test suites, code review of a small diff, focused research | | Opus | `opus` | Large features, cross-module refactors, architecture and design, subtle bugs, review of a large or risky diff | | Fable | `fable` | The hardest work only: cross-repo architecture, root-cause debugging nobody has cracked, novel algorithms, high-ambiguity planning, any task Opus already failed | If the session cannot use a model (the Agent call rejects it), use the next tier down and say so. ## Prompt template A fresh subagent has none of this conversation. Include: goal, files in scope and out of scope, conventions (point at an example file), done criteria with a verify command, and "Do not commit. Report changed files and anything unverified." ## After the subagent returns (mandatory) 1. Read `git diff` / `git status`. 2. Re-run the verify command yourself. 3. Fix small issues directly; re-delegate one tier up or take over for big misses. 4. Tell the user which worker and model did the task and what you verified.
Wersje
| Repozytorium | Wynik | Licencja |
|---|---|---|
| Ta strona · joezhuo2/delegate-skills | 100 | MIT |
| FrancoStino/opencode-skills-collection | 100 | MIT |
| sickn33/agentic-awesome-skills | 100 | MIT |
| amelnagdy/delegate-skills | 100 | MIT |
Migawka bazy z 13.11.2026 · wersja paczki 6c87ade9baa6 (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
- 100kiro-delegate
Przekazuje samodzielne, ograniczone zadanie do Kiro CLI działającego na modelu Claude i uruchamia je bez interfejsu. - 100subagent-model-picker
Dobiera model (Haiku, Sonnet, Opus lub Fable) dla każdego subagenta Claude Code przed uruchomieniem, z uzasadnieniem kosztu i jakości, i przekazuje go narzędziu Agent. - 100gemini-delegate
Przekazuje ograniczone zadanie do Gemini CLI jako dodatkowego wykonawcę, uruchamianego bez interfejsu, np. wdrożenie, testy lub recenzja. - 100delegate-codex
Zleca samodzielne prace agentowi Codex CLI i dobiera poziom modelu do trudności zadania, żeby oszczędzać limity Claude. - 100use-fable-subagent
Uruchamia prawdziwego subagenta Claude Fable do jednej analizy lub przeglądu tylko do odczytu, bez podmiany modelu i zapisu w repozytorium. - 100model-picker
Dobiera model Claude i poziom wysiłku do zadania, podagenta lub integracji API, a także planuje tańsze przejścia między modelami.