#82 Refactor: split fleet skill out of PM skill (#66)

[Anonymous]

Originally created by: kumaakh

Summary

• Fleet/PM skill boundary enforced: fleet owns infrastructure mechanics, PM owns orchestration — no cross-contamination
• PM autonomous workflow improvements: sprint selection gate, pre-flight checks, documentation harvest, backlog carried across all sprint types, BL-N IDs
• Background Agent rules clarified: execute_prompt always background (fleet hard rule); execute_command background if long-running; PM batches sequential ops into one Agent
• Reviewer dispatch improvements: always model=premium, never resume=true, SHA verification before dispatch
• Factual verification bullet added to tpl-reviewer.md — reviewers must spot-check URLs, package names, install commands
• tpl-backlog.md template created with BL-N identifiers, full-detail format, and source tracking
• Memory cleanup: 19 stale memory files deleted after confirming gaps addressed in refactored skill

Issues addressed

• Closes [#66] — fleet/PM skill split complete; fleet = infrastructure, PM = orchestration
• Closes [#67] — .fleet-task* added to member .gitignore in onboarding Step 7
• Closes [#48] — plan revision iterations use resume=true to preserve planning context
• Closes [#47] — reviewer dispatches always use model="premium"
• Closes [#44] — doer never resumes after phase boundary; fresh session per phase
• Closes [#41] — PM uses execute_command via fleet tools; bash-only rule documented in fleet Dispatch Rules
• Partially addresses [#73] — bash syntax preference documented (Git Bash universally available on dev machines)
• Partially addresses [#54] — fleet/PM boundary enforced; PM orchestration patterns consolidated

Changes by file

Fleet skill

• skills/fleet/SKILL.md — Dispatch Rules split into 3 explicit bullets (execute_prompt, execute_command, send/receive); bash-only rule with rationale
• skills/fleet/onboarding.md — Step 7: add .fleet-task* to member .gitignore

PM skill

• skills/pm/SKILL.md — Sprint Selection gate; Rule 7 clarified (club sequential fleet ops into one background Agent); status.md updated on every dispatch, not just phase boundaries
• skills/pm/single-pair-sprint.md (renamed from sprint.md) — branch naming, task harness sends project docs, documentation harvest before cleanup, recovery triage table
• skills/pm/simple-sprint.md — backlog update step, branch naming, recovery section, stale ref fixed
• skills/pm/multi-pair-sprint.md — backlog update step, contracts section, stale ref fixed
• skills/pm/cleanup.md — requirements.md design.md added to git rm (post documentation harvest)
• skills/pm/doer-reviewer.md — doer annotates feedback.md with **Doer:** fixed in commit <sha>; reviewer owns full content
• skills/pm/context-file.md — {sprint/skill-refactor} / {{base_branch}} placeholder fill before delivery
• skills/pm/tpl-doer.md — {sprint/skill-refactor} / {{base_branch}} placeholders; fast tests after every commit; full suite at VERIFY
• skills/pm/tpl-reviewer.md — cumulative scope clarified; git history check on re-review; factual verification bullet; feedback.md format template embedded
• skills/pm/tpl-reviewer-plan.md — git history check; feedback.md format template embedded; narrative required
• skills/pm/tpl-plan.md — Tier: cheap | standard | premium added to every task entry
• skills/pm/tpl-backlog.md — NEW: backlog item template with BL-N IDs, impact, detail, source
• skills/pm/init.md — backlog.md references tpl-backlog.md and BL-N identifiers

🤖 Generated with Claude Code

[Anonymous]

Ticket changed by: kumaakh

• status: open --> closed