crew-start
Reprend le travail crew sans préciser quoi — continue la tâche en CURRENT_TASKS, sinon démarre un batch inactif depuis TODO (anti-collision), puis code.
Skill by Nakken13
Chalk Labs
Need help deploying crew-start?
Chalk Labs sets up agents like this inside real teams: connected to your tools, tested on your data, with a human in the loop where it matters.
Install
Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.
npx agentmarket install skill/nakken13-crew-start
Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.
Get the weekly diff
The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.
About
Use cases: productivity
Reviews
Write a reviewNo reviews yet. Installed it? Reviews come from verified installers only. How reviews work
Permissions
What this skill can do and what it asks for, from its SKILL.md and code. Checked Nov 1, 2026.
- Tools
- No tools found
- Secrets
- None asked for
- Install scripts
- None found
Network hosts it talks to 1
codeload.github.com· found in install command
Files it touches
~/.claude/skills/crew-start
What's inside
- Skill name
crew-start- Description
- Reprend le travail crew sans préciser quoi — continue la tâche en CURRENT_TASKS, sinon démarre un batch inactif depuis TODO (anti-collision), puis code. Trigger — "/crew-start", "continue le travail", "lance la suite".
- License
- MIT
- Instructions
- 86 lines in SKILL.md · view on GitHub
Agent Skills spec issues
- files: referenced but missing: scripts/crew_hook.py
Files (1)
- SKILL.md
SKILL.md (first lines)
Single entry point to resume a crew session without naming a task. The task
**choice** is made by a deterministic script (`crew_hook.py next-task`); this
skill then runs the start (lock, worktree, `git mv`), the implementation, then
`/crew-close-task`.
**Repo-root anchoring is mandatory**: every `crew/...` path is resolved from the
repo root (`git rev-parse --show-toplevel`), never from the current cwd
(otherwise a phantom `crew/` is created in a subfolder). The
`git worktree add`/`cd` of steps 3-4 runs from the root of the main checkout.
## Steps
1. **Choose the task** (read-only, ~0 reasoning tokens):
```bash
ROOT="$(git rev-parse --show-toplevel)"
HOOK="${CLAUDE_PLUGIN_ROOT:+$CLAUDE_PLUGIN_ROOT/scripts}"; HOOK="${HOOK:-$ROOT/crew}/crew_hook.py"
CLAUDE_PROJECT_DIR="$ROOT" python "$HOOK" next-task
```
(plugin: the plugin's `scripts/crew_hook.py`; manual clone:
`crew/crew_hook.py`). JSON output: `action`, `task`, `batch`,
`worktree`, `branch`, `skipped[{slug,reason}]`, `paused`,
`unclassified`, `ambiguous`. It already covers: other sessions' locks,
active/paused batches, `Zone :` overlap, ICEBOX and `Unsorted` (to be
sorted) ignored. Do not redo this sorting by hand.
If `paused` is not empty, flag it to the user (task + what blocks it) — a
`PAUSED/` task is NEVER resumed automatically.
2. **According to `action`**:
- `resume` → step 3 (resume case). If `ambiguous`: ask the user which one to
prioritize rather than guess.
- `start` → step 3 (start case).
- `classify` → `TODO/` tasks have neither batch nor zone (`unclassified`):
dispatch `Agent({subagent_type: "manager"})` to classify them (impact zone +
batch, see `crew/CLAUDE_CONTEXT/LIFECYCLE.md` § Batching), then rerun
step 1.
- `none` → nothing startable: tell the user why from `skipped` (lock, active
batch, overlap) or "TODO empty", and stop. The only case where the skill
stops without having coded.
2bis. **Remote claim** (opt-in: only if `crew/CLAUDE_CONTEXT/CREW_CONFIG.md`
contains `remote_claims : true`, otherwise skip; claims are a layer shared
between devs on the `refs/crew/claims` ref, see `AGENTS.md`):
- if `action: start` and the task does not exist locally but is listed in
`remote_todo`: `python "$HOOK" claims materialize <task>` (writes the `.md`
into `crew/TODO/` from the ref);
- set the claim before any move:
`python "$HOOK" claims claim <task> --session "$CLAUDE_CODE_SESSION_ID" --batch "<batch>" --branch "<branch>" --zone "<zone from the JSON, comma-separated>"`;
- **on resume (`action: resume`) redo this same claim**: it refreshes `ts`
(the TTL is 8 h, a long task would otherwise become claimable again) and
publishes a task started offline;
- `ok: false` (already taken, overlapping zone, conflict) → do not insist:
rerun step 1 (the script sees the claim and suggests the next one);
- `offline` / `claims_warning` → **non-blocking**: continue and flag it to the
user ("remote claims not verified").
3. **Lock + isolation** (before any edit):
- *Resume*: run `Bash({command: ': "crew-resume:<task>"'})` (a no-op that sets
the lock through the `PreToolUse` hook).
- Create or reuse the batch worktree:
`git worktree add -b <branch> <worktree> main` (values from the JSON;
`git worktree list` first: if it already exists, do not recreate it,
including under an old `…-batch-<slug>-…` slug). `cd` into it and use
absolute paths of this copy for every following `Read`/`Edit`/`Write`.
- *Start*: from the worktree,
`git mv crew/TODO/<task> crew/CURRENT_TASKS/<task>`, update both
`INDEX.md` files.
- *Name the session* (start only; on resume the SessionStart hook already did
it): show the user `/rename <task slug without .md>` — a hook cannot rename
a running session, only SessionStart can.
4. **The hook blocks (exit 2)** — lock taken or batch busy between the choice and
the claim (race between sessions): do not insist. Rerun step 1 (the script now
sees the lock and suggests the next task). Only stop on `action: none`.
5. **Implement**: reread the task file, work through the `- [ ]` actions in
order, with the skills from the `CLAUDE.md` routing
(`test-driven-development`, `systematic-debugging`, `graphify`…). Tick each
action when it is really finished (not ahead of time, not in bulk).
6. All actions ticked → chain straight into `crew-close-task` (without asking
for confirmation, it is part of the protocol).
## What this skill does not do
- Never starts two tasks from different batches in the same invocation — one
`/crew-start` = one task until its closure (or a real blocker).
- Never picks from `crew/ICEBOX/` nor from `Unsorted` without
classification by `manager`.
- Never automatically resumes a task from `crew/PAUSED/`.
- Does not tick unfinished actions, does not skip the lock/anti-collision.
- Does not replace `/crew-close-task` — calls it at the end of the task.From Nakken13/claude-crew, MIT.
More from Nakken13/claude-crew
- 0.601score
crew-new-task
Nakken13
Crée une tâche crew (TODO ou CURRENT_TASKS) avec fichier, INDEX.md et batching via le persona manager.
SkillAFreshness: A★ 3Claude Code +9 - 0.601score
crew-close-task
Nakken13
Clôture une tâche crew terminée — vérifie les cases, passes review + simplify, historise, sort les tests IA/DEV, barre la tâche du batch.
SkillAFreshness: A★ 3Claude Code +9 - 0.597score
crew-update
Nakken13
Met à jour les fichiers moteur d'un projet bootstrapé vers la dernière version du scaffold, sans toucher aux données ni écraser un fichier personnalisé.
SkillAFreshness: A★ 3Claude Code +9 - 0.597score
crew-status
Nakken13
Rapport lecture seule de l'état crew — batchs actifs et zones, tâches en cours, tests IA non cochés, TODO non classées.
SkillAFreshness: A★ 3Claude Code +9 - 0.597score
crew-dashboard
Nakken13
Lance un dashboard web local temps réel (tâches, batchs, sessions concurrentes) avec actions de gestion.
SkillAFreshness: A★ 3Claude Code +9 - 0.597score
crew-count
Nakken13
Rapport lecture seule — combien de batchs de CLAUDE_BATCH.md sont lançables en parallèle (actifs, lançables, exclus pour chevauchement).
SkillAFreshness: A★ 3Claude Code +9 - 0.593score
crew-version
Nakken13
Affiche la version de claude-crew installée (vX.X.X).
SkillAFreshness: A★ 3Claude Code +9 - 0.576score
crew-init
Nakken13
Bootstrap le scaffold crew (CLAUDE.md, AGENTS.md, crew/…) sur un projet neuf ou existant — copie, détecte la stack, résout les placeholders.
SkillAFreshness: A★ 3Claude Code +9
Signals
- Stars
- 3
- Forks
- 1
- Open issues
- 0
- Last commit (this folder)
- Oct 16, 2026 (26 days ago)
From GitHub (fetched Oct 24, 2026).
Details
- License
- MIT
- Repository
- github.com/Nakken13/claude-crew
- First seen
- Oct 24, 2026
- Last checked
- Oct 24, 2026
- Source
- GitHubgithub-graphql
Same skill elsewhere
Copies of this SKILL.md published by others; search shows only one of them.
More from Nakken13
- 0.601score
crew-new-task
Nakken13
Crée une tâche crew (TODO ou CURRENT_TASKS) avec fichier, INDEX.md et batching via le persona manager.
SkillAFreshness: A★ 3Claude Code +9 - 0.601score
crew-close-task
Nakken13
Clôture une tâche crew terminée — vérifie les cases, passes review + simplify, historise, sort les tests IA/DEV, barre la tâche du batch.
SkillAFreshness: A★ 3Claude Code +9 - 0.597score
crew-update
Nakken13
Met à jour les fichiers moteur d'un projet bootstrapé vers la dernière version du scaffold, sans toucher aux données ni écraser un fichier personnalisé.
SkillAFreshness: A★ 3Claude Code +9 - 0.597score
crew-status
Nakken13
Rapport lecture seule de l'état crew — batchs actifs et zones, tâches en cours, tests IA non cochés, TODO non classées.
SkillAFreshness: A★ 3Claude Code +9 - 0.597score
crew-dashboard
Nakken13
Lance un dashboard web local temps réel (tâches, batchs, sessions concurrentes) avec actions de gestion.
SkillAFreshness: A★ 3Claude Code +9 - 0.597score
crew-count
Nakken13
Rapport lecture seule — combien de batchs de CLAUDE_BATCH.md sont lançables en parallèle (actifs, lançables, exclus pour chevauchement).
SkillAFreshness: A★ 3Claude Code +9
Done for you
Rather not set it up yourself?
Chalk Labs scopes, builds and runs agents for your team: the task audit, the eval harness, the integrations and the weekly ops report. Fixed scope, priced before we start.
See how we build AI agents →Read the case studies
Chalk Labs is a two-founder AI and growth studio. We build AI agents, CRMs and MVPs for Web3 and AI companies, and have shipped 50+ product and token launches.
- launches shipped
- 50+
- launches shipped
- media placements
- 500+
- media placements
- founders on every account
- 2
- founders on every account
Press we've placed: Forbes · CoinDesk · Decrypt · Cointelegraph · The Block
Rahil Jain, Co-founder, Chalk Labs. Ex-Goldman Sachs engineer. You talk to the person who builds it.
Get crew-start running in your business
Tell us what you want to automate. A founder replies within one working day with a plan, not a pitch deck.
Cite this page
crew-start skill. Agent Mart. https://agentmart.dev/skills/nakken13-crew-start (last updated 2026-10-24)