القائمة
Use after delegate_task leaves finish. Blind-compare candidate evidence, merge one coherent implementation, and avoid Frankenstein patches before verification.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
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-synthesize-winner |
| description | Use after delegate_task leaves finish. Blind-compare candidate evidence, merge one coherent implementation, and avoid Frankenstein patches before verification. |
| version | 0.1.0 |
| author | CodeGraphTheory |
| license | MIT |
| metadata | {"hermes":{"tags":["heavy-coder","synthesis","critique","swarm"],"related_skills":["heavy-team-default","heavy-leaf-brief","heavy-ship-gate","heavy-leaf-candidate-output"]}} |
Parallel leaves produce competing partial truths. The coordinator must rank evidence, pick a spine implementation, and port only justified hunks from runners-up. This skill is the post-swarm merge discipline between critique_candidates.py and final verification.
delegate_task results returned (or batch partially failed)patch / write_file on the main treeDo not use for:
/agents or swarm_watch.py)Collect - Gather each leaf summary; extract file lists, test commands, exit codes, SHAs if any.
Validate JSON (when candidates emit structured output):
python skills/heavy-issue-to-merge/scripts/validate_candidate.py path/to/candidate.json
Critique (deterministic rank when JSON available):
python scripts/critique_candidates.py --candidates dir/with/json --repo .
Blind compare - Score on: test evidence > scope discipline > regression risk > maintainability. Ignore eloquence.
Choose spine - One candidate (or one approach) owns the base diff.
Cherry-pick hunks - Import only hunks with independent evidence (test or cited line-level reason).
Re-read - read_file every file you will touch after mental merge.
Apply - patch / write_file once on the integration branch or main worktree.
Hand off - Run heavy-ship-gate / plan verification; do not claim done here.
Done when: working tree reflects one story, not a collage of conflicting designs.
Profile hooks may inject DELEGATE_TASKS_JSON. After leaves return, synthesis is still coordinator responsibility; hooks do not auto-merge code.
critique_candidates.py when JSON exists.candidate_id, files, and commands for auditability.Output to the user (before ship gate):
Load heavy-ship-gate for the actual test/lint execution and completion message.