Menú
Orchestrate multi-sprint phases where arch-ctm (Codex) is the sole developer, with pipelined QA via quality-mgr teammate. Team-lead tracks findings and schedules fix passes.
Instalar con Codex o Claude Copia este prompt, pégalo en Codex, Claude u otro asistente, y deja que revise la página de la skill y la instale por ti.
Basado en la clasificación ocupacional SOC
Generate a sprint status report for the current phase. Default is --table.
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.
Reusable QA orchestration skill for GitHub PRs. Use for multi-pass QA, CI monitoring with `atm gh monitor`, one-shot PR reporting with `atm gh pr report`, and template-driven findings/final quality reports.
Orchestrate multi-sprint phase execution as team-lead (ARCH-ATM). Manages sprint waves, scrum-master lifecycle, PR merges, arch-ctm reviews, and integration branch strategy. This skill is for the TEAM-LEAD only, not for scrum-masters.
Launch and verify named teammates that participate in team messaging. Use when setting up a new team member with mailbox polling, troubleshooting why a teammate is not receiving messages, or creating named Claude/Codex/Gemini teammates with team-scoped communication.
Run ATM health diagnostics, present severity-sorted findings, and delegate deep remediation analysis to the atm-doctor agent when critical findings exist.
| name | codex-orchestration |
| description | Orchestrate multi-sprint phases where arch-ctm (Codex) is the sole developer, with pipelined QA via quality-mgr teammate. Team-lead tracks findings and schedules fix passes. |
This skill defines how the team-lead (ARCH-ATM) orchestrates phases where arch-ctm (Codex) is the sole developer, executing sprints sequentially while QA runs in parallel via a dedicated quality-mgr teammate.
Audience: Team-lead only.
When to use: When a phase's implementation is done entirely by arch-ctm (a Codex agent communicating via ATM CLI), not by Claude Code scrum-masters. This pattern was proven in Phase M (8 sprints) and Phase O.
Before starting a phase:
integrate/phase-{P} created off developatm-dev) is active with team-lead and arch-ctm as membersatm send arch-ctm "ping")team-lead (ARCH-ATM)
├── arch-ctm (Codex) ──── sole developer, sequential sprints
│ communicates via ATM CLI only
└── quality-mgr (Claude Code) ──── QA coordinator teammate
spawns rust-qa-agent + atm-qa-agent as background agents
Key principle: arch-ctm does NOT wait for QA. He proceeds to the next sprint as soon as he completes one, unless there are outstanding fix requests from earlier sprints.
git fetch origin develop
git branch integrate/phase-{P} origin/develop
git push -u origin integrate/phase-{P}
# Use sc-git-worktree skill
/sc-git-worktree --create feature/p{P}-s1-{slug} integrate/phase-{P}
Spawn once per phase. The quality-mgr persists across all sprints.
Use the Task tool with name parameter to spawn as a named teammate:
{
"subagent_type": "quality-mgr",
"name": "quality-mgr",
"team_name": "atm-dev",
"model": "sonnet",
"prompt": "You are quality-mgr for Phase {P}. You will receive QA assignments from team-lead for each sprint as they complete. Stand by for first assignment. Integration branch: integrate/phase-{P}. Phase docs: docs/project-plan.md, docs/atm-agent-mcp/requirements.md."
}
atm send arch-ctm "Phase {P} Sprint {P}.1 assignment: {title}
Worktree: /path/to/worktree
Branch: feature/p{P}-s1-{slug}
PR target: integrate/phase-{P}
Deliverables:
- {list deliverables}
Requirements: docs/atm-agent-mcp/requirements.md ({relevant FRs})
Sprint plan: docs/project-plan.md (Phase {P} section)
When complete: commit, push, then notify me via atm send with branch + commit SHA. Do not create PR."
Timeline:
arch-ctm: [── S.1 ──]──fixes──[── S.2 ──]──fixes──[── S.3 ──]
quality-mgr: [── QA S.1 ──] [── QA S.2 ──] [── QA S.3 ──]
team-lead: assign S.1 → track → assign S.2 → track → assign S.3 → track
integrate/phase-{P} and immediately starts CI monitoring:
atm gh monitor pr <PR_NUMBER>
/sc-git-worktree --create feature/p{P}-s{N+1}-{slug} feature/p{P}-s{N}-{slug}
"Run QA on Sprint {P}.{S}. Worktree: {path}. Sprint deliverables: {summary}.
Design docs: {list}. PR: #{N}."
Priority order for arch-ctm:
Fix workflow:
# arch-ctm fixes on the sprint's original worktree
# arch-ctm pushes fix commits to same PR branch
# team-lead asks quality-mgr to re-run QA on the fixed worktree
# If QA passes, team-lead merges PR to integration branch
After fixes merge to integrate/phase-{P}:
git fetch origin
git merge origin/integrate/phase-{P}
Assignment format:
Run QA on Sprint {P}.{S}: {title}
Worktree: {absolute path}
Sprint deliverables: {bullet list}
Design docs: {list of relevant doc paths}
PR: #{number}
Re-run after fixes:
Re-run QA on Sprint {P}.{S} (post-fix).
Worktree: {path}
Fixed findings: {list of QA IDs addressed}
quality-mgr reports PASS/FAIL with finding IDs. Team-lead tracks:
| Sprint | QA Run | Verdict | Blocking Findings | Status |
|---|---|---|---|---|
| O.1 | 1 | FAIL | QA-001, QA-002 | Fixes assigned |
| O.1 | 2 | PASS | — | Merged |
| O.2 | 1 | PASS | — | Merged |
OPEN → assigned to arch-ctm → FIXED (arch-ctm pushes) → re-QA → VERIFIED (QA passes)
→ WONTFIX (team-lead approves deviation)
integrate/phase-{P} (never develop directly)integrate/phase-{P} → developAll arch-ctm communication is via ATM CLI. Follow the dogfooding protocol (ACK → work → complete → ACK).
atm send arch-ctm "message"
atm read
Use ATM first:
atm send arch-ctm "You have unread ATM messages. Run: atm read --team atm-dev"
If your local runtime uses tmux pane orchestration, tmux nudges are optional and environment-specific.
When arch-ctm is waiting for assignments, tell him:
"Standing by? Use: atm read --team atm-dev --timeout 60"
This keeps him responsive without busy-polling.
After all sprints pass QA and merge to integration branch:
integrate/phase-{P} → developdevelop — chain from previous sprint or integration branch