القائمة
Load shared swarm behavioral rules for coordinators and workers. Use for task routing, receipts, direct messaging, dedup, worktree boundaries, and lifecycle discipline.
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
Load perl-lsp coding standards for code-writing or code-review work in this repo. Use when implementing, fixing, reviewing, or validating code.
Merge approved PRs in safe batches — checks CI, fixes policy failures, handles conflicts, ratchets corpus after parser merges.
Run a TDD parser-fix workflow for perl-parser-core or related parser crates. Use when a parser bug has a concrete failing construct and needs tests, a minimal fix, and verification.
Turn a discovery into a builder-ready handoff packet. Use when a scout or coordinator has identified a concrete slice and needs exact scope, verification, and routing rather than code changes.
Research-first workflow that scouts exact scope before launching a constrained builder. Use when a feature or fix is too unconstrained for a direct builder — converts ~50% success into ~90% by eliminating implicit decisions.
Load roadmap weighting for scouts and coordinators so slices are ranked by actual repo priorities instead of convenience.
| name | swarm-protocol |
| description | Load shared swarm behavioral rules for coordinators and workers. Use for task routing, receipts, direct messaging, dedup, worktree boundaries, and lifecycle discipline. |
| user-invocable | false |
Shared behavioral rules for swarm coordinators and workers.
Fix adjacent same-slice issues in the current worktree when they are clearly part of the task. For anything outside the slice, create a GitHub issue or handoff instead of stretching the worker.
Use swarm-discovered for discovered issues and swarm-architectural when the
repo needs a deliberate architecture decision.
Message the teammate who should act next. Do not bounce everything through the lead.
Use GitHub and tracked swarm state as the durable ledger:
.claude/swarm-state/Do not treat transcript memory as proof.
Use .claude/swarm-state/findings.json for durable control-plane conclusions.
Use discovered-issues.md for product leads and known-pitfalls.md for
repeatable failure lessons.
After each task, leave a durable receipt:
When the lane uses .ops-perl-lsp/swarm-metrics.jsonl, append an entry instead
of rewriting history.
Check the queue and state before creating new work:
.claude/swarm-state/If the slice already exists, update or route it instead of duplicating it.
If the work changes how the swarm should describe or operate itself, update the findings ledger instead of hiding that conclusion in chat or a one-off handoff.
Stop as proof of completionWhen facts matter, look them up or spawn the right research worker. Do not guess at language rules, protocol details, or dependency behavior.