Loading...
Loading...
Manage your AI team. /crew = roster, /crew add [name] "[domain]" = new member, /crew [name] [task] = assign task.
npx skill4agent add eran-tookii/crew crew| Arguments | Action |
|---|---|
| none | Show roster |
| Scaffold a new crew member |
| Activate member, ask for task |
| Activate member with task |
.claude/team/*/SKILL.md.claude/team/SKILL.mdnamedescriptionhistory.md/crew/crew [name]/crew [name] [task]/crew add [name] "[domain]"add [name] "[domain]".claude/team/{name}/already exists as a crew member. Overwrite? This will replace{name},SKILL.md,context.md, andhistory.md— all existing context and history will be lost.decisions.md
/crew {name}.claude/team/{name}/SKILL.md---
name: {NAME}
description: Domain expert for {DOMAIN}.
user-invocable: false
---
# {NAME} — {DOMAIN}
## On every task — before writing a single line of code
1. Read `.claude/team/{name}/context.md` — understand the current state of the domain
2. Read `.claude/team/{name}/decisions.md` — key architectural decisions, gotchas, things not to undo
3. Read the last entries in `.claude/team/{name}/history.md` — recent work and what just changed
4. Read any relevant skill files for this domain (best practices, frameworks, etc.)
5. Only then proceed with the task
## On every task — after completing the work
1. **Update `context.md`** — reflect any changes to files, patterns, architecture, or configuration
2. **Append to `history.md`** — add a dated entry with:
- What was done and why
- Which files changed
- Run `git log --oneline -1` and record the commit ID and message
- PR number if one exists
- Any decisions made and their rationale
- Any gotchas discovered
3. **Maintain the history window** — if `history.md` has more than 10 entries:
- Extract any important decisions, gotchas, or patterns into `decisions.md` before they are lost
- Trim `history.md` to the 10 most recent entries
4. **If this task produced a key architectural decision or surfaced a non-obvious gotcha** — add it to `decisions.md` directly, even if no trimming is needed
5. **Sign off** — end your response with a single line: `— {name}, {domain}`.claude/team/{name}/context.md# {NAME} — Domain Context: {DOMAIN}
Last updated: {TODAY'S DATE}
## What This Domain Does
[Describe what this domain is responsible for — 2-3 sentences]
---
## Key Files
| File | Role |
|------|------|
| `path/to/file` | What it does |
---
## Architecture
[Describe the architecture — how data flows, what calls what]
---
## Key Patterns
[Document the patterns used in this domain]
---
## Important Constraints & Gotchas
[Things that are non-obvious, decisions that look wrong but aren't, things not to change]
---
## Environment Variables
[Any env vars this domain depends on]
---
## Dependencies
[Key packages this domain uses].claude/team/{name}/history.md# {NAME} — Domain History: {DOMAIN}
Capped at 10 entries. When trimmed, important decisions and gotchas are extracted to `decisions.md` first.
---
## {TODAY'S DATE} — Initial setup
**Added by:** /crew add
Domain expert created for: {DOMAIN}
Fill in `context.md` with the current state of the domain before assigning the first task..claude/team/{name}/decisions.md# {NAME} — Decisions & Institutional Memory: {DOMAIN}
Evergreen knowledge extracted from history: architectural decisions, known gotchas, and things not to undo. Updated when important decisions are made or when history.md is trimmed.
---
## Key Decisions
[Record important architectural choices and the reasoning behind them]
---
## Known Gotchas
[Non-obvious behaviors, traps, and things that look wrong but aren't]
---
## Things Not To Change
[Deliberate decisions that might look like bugs or bad patterns — with the reason why]/crew {name} — look into the project and find areas relevant to your domain '{DOMAIN}'. Review the codebase and any relevant documentation, then create your context.md based on what you find.
[name][name] [task].claude/team/{name}/SKILL.md— {name}, {domain}