Menu
Use when the user says "new epic", "new task", "new standalone", "end session", "update status", "handover", "handover-resume
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Basé sur la classification professionnelle SOC
Panel-only CR gate for Codex — runs the shell critic panel and CodeRabbit CLI pass over the branch diff, records ledger evidence, and clears the CR marker through the sanctioned chokepoint only when both the panel and CodeRabbit findings are clean (no Critical/Important blockers). Use when the user asks to run /pr-check or clear the CR gate under Codex.
Arm the OS scheduler to relaunch claude at a given time with a given handover. Dedup-guarded. Use when the user asks to arm a resume / schedule a relaunch or run /handover-arm-resume.
Arm/inspect/remove the recurring luna clip-pipeline cadence (daily harvest+triage, daily synthesize+archive, daily health) via schtasks/cron, each leg pinned to a cheap --model. Dedup-guarded. Use when the user asks to arm/check/disarm the clip-pipeline cadence or runs /pipeline-cadence.
Use when the per-project auto-memory store (`~/.claude/projects/<project-slug>/memory/`: the always-loaded `MEMORY.md` index + one topic file per fact) is approaching or over its load budget (~24.4KB) — the harness emits a size-limit warning at session load when over budget (e.g. "MEMORY.md is NN KB (limit 24.4KB)"), content past the limit is silently dropped = partial recall. Losslessly compounds the DURABLE gotchas out of the auto-memory into qmd-searchable luna / himmel reference notes (read-many → propagation review → write-once → qmd gate → slim index → delete sources), so the always-loaded index stays lean and adopter-generic learnings land in himmel docs (memory + vault never propagate to other users). Lean-invoke, operator-run on demand — NOT an always-on hook (HIMMEL-569).
Use when turning an idea into an implementation plan — runs brainstorm → spec → plan as ONE pipeline with an ADVERSARIAL CRITIC LOOP between each stage. Trigger on "build/implement/design X", "/minerva", or any feature/capability work that should pass through a spec + plan before code. Composes superpowers brainstorming + writing-plans (no fork); adds a spec-critic and a plan-critic.
Decision guide for delegating to caveman-style subagents. Tells the main thread WHEN to spawn `cavecrew-investigator` (locate code), `cavecrew-builder` (1-2 file edit), or `cavecrew-reviewer` (diff review) instead of doing the work inline or using vanilla `Explore`. Subagent output is caveman-compressed so the tool-result injected back into main context is ~60% smaller — main context lasts longer across long sessions. Trigger: "delegate to subagent", "use cavecrew", "spawn investigator/builder/reviewer", "save context", "compressed agent output".
| name | handover |
| description | Use when the user says "new epic", "new task", "new standalone", "end session", "update status", "handover", "handover-resume |
Tracks epics, tasks, and standalones across sessions. Multi-repo capable: state lives per registered repo under <repo-root>/handovers/<user>/, optionally bucketed by source repo under <repo-root>/handovers/<user>/<bucket>/ (HIMMEL-129 layout — see references/resolution.md).
This SKILL.md is a thin router (HIMMEL-1041). It carries only the op index + invariant rules. Each op's mechanics live in references/<op>.md, loaded on demand — so invoking one op never loads the other ~19. The read path (handover-resume) is fully script-driven and skips the skill entirely (HIMMEL-1038).
All mutation ops resolve the target repo the same way and share the registry protocol, template placeholders, the No-ID picker, status values, and the file-path map — that shared substrate lives in references/resolution.md; load it first, then the op's own slice below. Most (new-epic, new-task, new-standalone, end-session, update-status, bucket, priority, jira-link, hygiene, consolidate) also resolve bucket and run inside the worktree gate before writing; next-ID derivation applies only to the new-* creation ops. defaults is the exception — it resolves only the target repo and writes ~/.claude/handover/registry.json (no bucket, ID, or worktree). Read-only ops (handover-resume, repos list) skip the worktree gate.
new-epic <name> · new-task <epic-id> <name> · new-standalone <name> — create a tracked item (Jira auto-create gate + worktree + templates). → references/new-item.mdupdate-status — regenerate status.md + roadmap.md + tech-debt.md from filesystem state; run after any mutation, never ask. → references/update-status.mdend-session [id] — write the next append-only next-session-N.md (session summary + cold-start prompt + Jira transition); carries the overnight-mode trigger. → references/end-session.md/handover bucket <id> <bucket> (time-horizon OR source-repo move) · /handover priority <id> <priority> · /handover jira-link <id> [<key>] · /handover defaults <sub> → references/item-ops.md/handover bug <add|fix|status> · /handover bugs [--open] · /handover lessons → references/bug-ops.md/handover hygiene [triage|consolidate|analyse] · /handover consolidate apply <N> → references/hygiene.mdhandover-resume #N — the /handover-resume command calls scripts/handover/resume.sh directly and does NOT load this skill (HIMMEL-1038). If reached via skill trigger: → references/resume.md/handover-setup — first-time bootstrap (Mode A inline / Mode B external state repo). Full runbook: the plugin's commands/handover-setup.md./handover init — bootstrap a NEW target repo (no existing state). → references/init-register.md/handover register — adopt EXISTING state, idempotent. → references/init-register.md/handover repos <list|add|remove|where> — manage the repo registry (read-only reads; list skips the worktree gate). → references/init-register.md<state-root>. Bucket layer = any recognized source bucket exists directly under <state-root> — the four built-ins himmel/, luna/, luna_brain/, cross/ plus any names in the host repo's source_buckets_extra (HIMMEL-307). See references/resolution.md.counter.md alone — always reconcile against filesystem max.<state-root> happen in a worktree of the target repo — never on main. The machine-global registry (~/.claude/handover/registry.json, written only by defaults/repos) is not a target-repo write, so it needs no worktree (see the op index above).<state-root>/_templates/ files via mutation commands. Copy and fill.update-status.# prefix is part of dir name.When user gives feedback on work during chat, capture it to the relevant reviewer-notes.md under ## Human Feedback. Proactively — don't wait to be asked. Resolves under the target repo of the work being reviewed.
When a fix is attempted during debugging and fails, append it to the active item's bug before moving on — proactively, don't wait to be asked: /handover bug fix <BUG-n> FAILED "<what you tried + why it failed>" (or bug add "<symptom>" first if the bug isn't tracked yet). When a fix works, record WORKED and set status resolved. This Fixes tried FAILED/WORKED ledger is the circular-debugging breaker — it stops a later session (or a post-compaction you) from re-trying a fix that already failed. Resolves under the target repo of the work being debugged (same resolver as Human Feedback).
references/resolution.md — shared substrate: repo/bucket/registry resolution, ID derivation, worktree gate, template placeholders, No-ID picker, status values, supplementary files, file-path mapreferences/new-item.md — new-epic / new-task / new-standalonereferences/update-status.md — update-status + roadmap.md / tech-debt.md generationreferences/end-session.md — end-session + cold-start format + overnight modereferences/item-ops.md — bucket / priority / jira-link / defaultsreferences/bug-ops.md — bug / bugs / lessonsreferences/resume.md — handover-resume (script-driven via scripts/handover/resume.sh)references/routing.md — full resolution algorithm, canonicalisation, edge casesreferences/init-register.md — full specs for init, register, repos; registry schema + defaults keysreferences/buckets.md — WIP context-detection rule + bucket vocabreferences/hygiene.md — stale tiers, lingering detection, triage/consolidate/analyse + hygiene/consolidate op flowreferences/sync.md — bidirectional Jira syncreferences/v2-schema.md — v2 frontmatter schema