메뉴
Close a wave deterministically — run the gate prerequisites, dispatch the fresh-context adversarial review, and record the close. Use once per wave when the wave's bricks are shipped and validate is green.
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
In-loop development guidance for an autonomous /goal Lean-proof loop in SK_EFT_Hawking. This skill should be used when developing Lean 4 proofs inside a goal loop, deciding whether to fan work out to parallel worktree slots or stay solo, resetting a worktree slot, hitting a recurring Lean tactic/ elaboration friction, or sharding the lab notebook. It surfaces the MCP-first proof loop, kernel-purity rules, the worktree fan-out flow, and a symptom-indexed friction catalog. Invoke it at the start of a development stretch and whenever you hit proof-mechanics friction or need the parallel-dev recipe.
Goal-mode LAUNCH + always-on posture for an autonomous /goal dev loop. On invocation it composes the /goal condition AND registers the session with the harness; its core posture re-attaches after compaction. Use when STARTING a managed /goal dev loop, or writing/refining a goal prompt + acceptance criteria. (For IN-LOOP development guidance — the MCP proof loop, worktree fan-out, friction catalog — use the goal-dev skill instead.)
Off-hot-loop System-2 harvest. Reads each managed /goal session's transcript from a byte-offset watermark, extracts process/harness signal (Haiku) — crossing the compact boundary for the pre-vs-post-compact delta and ingesting the guard's blocked-question log — and a register-AWARE Opus consolidator files/combines it into the four-section register (Open / Process Wins / Closed / Misfiled) — re-opening a recurring closed finding, grouping semi-related ones, filing real wins to Process Wins (capped at agent-reviewed, never injected), and dropping noise (never writing Misfiled — that is /debrief's human sweep) — then refreshes the open-only active-issues view. Invoked by a Desktop scheduled task or a second-session /loop — never inside a /goal session.
Interactively review the System-2 dev-process harvest — promote agent-reviewed findings to human-reviewed (or close them), and triage GAP-A gate proposals. Runs over the register-aware harvest's already-synthesized register (it files/combines/re-opens continuously); debrief is the human governor — only it promotes to human-reviewed. Use when you want to sign off on what the autonomous loop learned about HOW it ran.
Run the full mechanical Stage-12 sync (counts, tables, deps, citation cache) in one command. Use before a wave gate, or whenever counts/tables/figures may have drifted from the Lean/Python sources.
| name | wave-close |
| description | Close a wave deterministically — run the gate prerequisites, dispatch the fresh-context adversarial review, and record the close. Use once per wave when the wave's bricks are shipped and validate is green. |
| argument-hint | <wave-or-bundle-id> |
| allowed-tools | Bash(git rev-parse *), Bash(cd *), Bash(uv run python *), Agent |
Close wave $ARGUMENTS in one deliberate pass. Do NOT do this per-task — once per wave.
Resolve the repo root (cwd-robust — works from the workspace root OR inside the repo) (NOT
${CLAUDE_PROJECT_DIR} — not a skill substitution):
!`R=$(uv run --no-sync python "${CLAUDE_PLUGIN_ROOT}/scripts/harness_common_cli.py" repo-root 2>/dev/null); test -n "$R" || R=$(git rev-parse --show-toplevel 2>/dev/null); echo "${R:-UNRESOLVED}"`. This uses the harness
repo_root() (the SAME resolver the hooks use), falling back to git rev-parse, so it resolves whether CC was
launched from the workspace root (the lean-lsp launch point) or inside the repo. UNRESOLVED only if
launched entirely outside the workspace — then tell the user to cd into SK_EFT_Hawking/ and re-run. Use
<repo-root> for the paths below.
Full sync + gate prereqs (deterministic). Run via Bash (the scripts self-locate their own ROOT;
sync.py's shared-artifact regen is serialized by the Task-7 regen concurrency lock, so a concurrent
worktree/worker can't race the same regen — spec 12):
cd "<repo-root>" && uv run python scripts/sync.py --fullcd "<repo-root>" && uv run python scripts/gate_precheck.py s13
(cd into the resolved repo so uv finds its project — a bare uv run from the workspace root runs an ephemeral interpreter WITHOUT the repo's deps (e.g. src.core fails to import); cd into the repo gives it the project env.) If gate_precheck s13 is FAIL, STOP — fix the red checks first; do not spend an
Opus review on a failing tree.Dispatch the fresh-context review (a real model call, the GAP-B safeguard the
self-audit blind spot requires): invoke the skeft-qa:adversarial-reviewer
(and skeft-qa:claims-reviewer for paper-shaped output) on $ARGUMENTS in a
fresh context. It must run with zero BLOCKER findings surfaced in the
transcript before the wave can be called closed.
Record the close. Write docs/dev-loops/<roadmap>/<wave>_close.md with: the
validate result, the reviewer report path, the BLOCKER count (must be 0), and the
commit range. Never edit CLAUDE.md / hooks / roadmaps here.
State plainly whether the wave is closeable (zero BLOCKERs, green tree) or what
remains. This is the explicit complement to the /goal acceptance-criteria path
(the goal-prompt skill bakes the same zero-BLOCKER requirement into the loop).