메뉴
After failed candidates or a red ship gate, spawn a narrow focused delegate_task wave with failure evidence instead of re-running a full-width council.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Keep coordinator and leaf context lean-batch reads, cap logs, use execute_code to filter-so swarms stay fast and within token limits.
Use at the start of non-trivial software tasks. Gather repo ground truth with batched reads and searches before planning or delegate_task.
Obey Plan 1A hook session phases so mutating tools, delegate_task sizing, and re-planning do not waste turns or trigger blocks.
Use when spawning delegate_task leaves for coding work. Pack self-contained goals and context so parallel Composer workers ship verifiable patches without coordinator chat history.
Require structured candidate-result JSON from swarm leaves so critique_candidates.py and synthesis run on evidence, not prose self-reports.
Enrich slim hook-injected delegate_tasks with touch map, failure excerpts, and verification commands so parallel leaves produce verifiable patches without coordinator chat history.
| name | heavy-repair-wave |
| description | After failed candidates or a red ship gate, spawn a narrow focused delegate_task wave with failure evidence instead of re-running a full-width council. |
| version | 0.1.0 |
| author | CodeGraphTheory |
| license | MIT |
| metadata | {"hermes":{"tags":["heavy-coder","delegation","debugging","performance"],"related_skills":["heavy-synthesize-winner","heavy-ship-gate","heavy-leaf-brief","systematic-debugging"]}} |
Re-running eight leaves on the same failing test burns tokens without new information. A repair wave is 1-3 focused delegate_task goals with the failing command output, current diff, and a tight hypothesis-then ship gate again.
heavy-synthesize-winner produced a merge but heavy-ship-gate exit code ≠ 0Do not use when:
heavy-explore-first + systematic-debugging firstReproduce on coordinator tree - Run the failing command; capture exit code and ≤80 lines (heavy-context-budget).
Freeze scope - One bug, one test file, or one API contract. State what is out of scope.
Design 1-3 goals (distinct angles, not full role rotation):
| Slot | Angle |
|---|---|
| A | Minimal fix for failing assertion only |
| B | Test-first: adjust test if spec was wrong |
| C | Adjacent call path / regression sibling |
Context packet - Include: repro command output, git diff --stat, touch map, evidence bar (single command), link to prior winner candidate_id if any.
Dispatch - delegate_task(tasks=[...]) with count ≤ 3 unless user requests wider repair.
Note: Hooks may still enforce minimum width from profile. If blocked, user may say single mode for coordinator-only fix, or coordinator fixes with evidence after explicit single-mode opt-in.
Synthesize - heavy-synthesize-winner on repair results only; do not merge unrelated prior candidates.
Ship - heavy-ship-gate must pass before "done".
Done when: failing command passes on coordinator tree or blocker is reported honestly.
Escalate to full council (8+) only when:
| Symptom | Fix |
|---|---|
| Same failure after repair | Deeper systematic-debugging; read definition sites |
| Full swarm again | Repair wave first |
| Ship without re-run | Ship gate on coordinator tree |