Menu
Session initialization for the team-lead identity. Confirms identity and detects whether a full team restore is needed. Only run when ATM_IDENTITY=team-lead.
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Basé sur la classification professionnelle SOC
Thin Claude skill wrapper for Synaptic Canvas package management through `sc --json`.
Team-lead delegates plan hardening to csc after the user has already discussed the plan details with csc.
Orchestrate multi-sprint phases where csc (Codex) is the sole developer, with pipelined QA via quality-mgr teammate. Team-lead tracks findings and schedules fix passes.
Create, manage, scan, update, and clean up git worktrees for parallel development with protected branch safeguards. Use when working on multiple branches simultaneously, isolating experiments, updating protected branches (main/develop), or when user mentions "worktree", "parallel branches", "feature isolation", "branch cleanup", "worktree status", or "update main/develop".
| name | team-lead |
| description | Session initialization for the team-lead identity. Confirms identity and detects whether a full team restore is needed. Only run when ATM_IDENTITY=team-lead. |
Trigger: Run at the start of every session where ATM_IDENTITY=team-lead.
echo "ATM_IDENTITY=$ATM_IDENTITY"
Stop if ATM_IDENTITY is not team-lead.
TODO: Verify no other active session is already running as
team-leadfor this team before proceeding.
Get the current session ID from the SessionStart hook output at the top of
context (format: SESSION_ID=<uuid>). Compare with leadSessionId in the
team config:
python3 -c "import json, os; team = os.environ['ATM_TEAM']; print(json.load(open(f'/Users/randlee/.claude/teams/{team}/config.json'))['leadSessionId'])"
docs/project-plan.md and outputting project status..claude/skills/team-lead/backup-and-restore-team.md.After initialization, the team-lead uses these skills to coordinate the team:
| Skill | Trigger |
|---|---|
/phase-orchestration | Orchestrate a multi-sprint phase (sprint waves, scrum-master lifecycle, integration branch, arch-ctm reviews) |
/codex-orchestration | Run phases where arch-ctm (Codex) is sole dev, with pipelined QA via quality-mgr |
/quality-management-gh | Multi-pass QA on GitHub PRs; CI monitoring; findings/final quality reports. Simple fixes/small features only — team-lead runs atm-qa-agent + rust-qa-agent directly in parallel. For multi-sprint phases use /phase-orchestration or /codex-orchestration instead. |
/sprint-report | Generate phase status table or detailed report |
/atm-doctor | Run ATM health diagnostics; escalate critical findings to atm-doctor agent |
/named-teammate-launch | Launch and verify named teammates (Claude/Codex/Gemini) with mailbox polling |
Additional orchestration guides are in
.claude/skills/*/SKILL.md. Consult the relevant skill before starting a new phase or delegating to a teammate.
⚠️ For any multi-sprint phased development,
/codex-orchestrationor/phase-orchestrationMUST be used as directed by the user. Using ad-hoc coordination instead of these skills leads to process drift, missed communications, and inconsistent QA gates.
After every session start or context compaction, if a phase is in progress:
/codex-orchestration or /phase-orchestration. Read only that one.Do not read both skills. Do not guess. If unsure — ask first, read immediately.
Skipping this re-read is the primary cause of process drift between sessions.
When assigning work to any teammate:
TaskCreate or TaskUpdate with assignee and description before sending the first message..claude/skills/team-lead/cross-platform-guidelines.md, Rust guidelines, etc.)/codex-orchestration skill) that require:
atm gh monitor pr <NUMBER>