wholework

Compare 3-panel comparison images (Before / After / Diff highlight) and enumerate visual gaps as structured JSON

Precedent Investigation: extract learnings from similar Issue/Spec retrospectives (for L/XL Issue parallel investigation)

Risk Investigation: assess test impact, verify command effects, and breaking change potential (for L/XL Issue parallel investigation)

Scope Investigation: identify change targets and impact scope (for L/XL Issue parallel investigation)

Recovery diagnostician for unknown orchestration failures. Analyzes phase state, wrapper exit code, and log tail to produce a minimal recovery plan in JSON format.

Detect documentation/implementation drift and auto-generate Issues (`/audit drift`), and detect structural fragility (`/audit fragility`). AI detects semantic drift between Steering Documents + Project Documents and codebase implementation, and auto-generates Issues for code-side fixes. Where `/doc sync` proposes documentation-side fixes, `/audit` is the complementary skill that creates Issues for code-side fixes. Running without arguments executes both drift and fragility perspectives in an integrated run. `/audit stats` aggregates Issue metadata across the project and generates a project health diagnostic report (throughput / composition / First-try success / Backlog Health, etc.), providing a third lens for project health alongside drift and fragility detection. `/audit stats --retention` adds phase/verify and Icebox dwell metrics (median/p95/30-day threshold violations, verify-type breakdown, Icebox dwell, trigger candidates) with escalation-based retire-proposal comment posting (30/60/90 days for verify, 90/180 days for Icebox). `/audit recoveries` reads the cross-Issue orchestration recovery log (`docs/reports/orchestration-recoveries.md`) and files Issues for recurring patterns that exceed a frequency threshold. `/audit progress <XL-parent-issue-number>` displays a sub-issue progress snapshot (status breakdown, phase distribution, time estimate, 24h activity) for the specified XL parent issue. `/audit auto-session <session-id>` generates the data layer of a /auto session retrospective report from `.tmp/auto-events.jsonl` filtered by session_id (SESSION_ID generated by /auto at startup as PID-timestamp). `/audit auto-session --full <session-id>` additionally generates LLM drafts for all 4 narrative sections (What worked / Limits and gaps / Improvement candidates surfaced / Conclusion) with `[LLM draft — human review required]` markers; no issues are auto-filed (human gate preserved). Both modes also generate a Japanese-translated sibling file at `{report-path-without-ext}-ja.md` by default; pass `--no-ja` to skip.

Autonomous execution (`/auto 123`). Runs spec (when needed)→code→review→merge→verify in sequence. XL Issues use sub-issue dependency graph with parallel execution. Size auto-detection with `--patch`/`--pr` and `--review=light`/`--review=full` overrides. Issues without `phase/*` labels start from issue triage. `--batch N` processes N backlog XS/S Issues; `--batch N1 N2 ...` processes the explicitly listed Issues in order (assigns a BATCH_ID for parallel-safe checkpointing). `--resume N` resumes a single Issue (restores verify counter from checkpoint); `--batch --resume` resumes an interrupted batch using `list_active_batches` to identify the target session.

Local implementation (`/code 123`). Size auto-detection routes XS/S→patch (direct commit to main), M/L→branch+PR. Override with `--patch`/`--pr`. Does not update CLAUDE.md, run session retrospectives, or manage memory.

Project foundation document management (`/doc init`). Centrally manages Steering Documents and Project Documents, ensuring consistency through drift detection and normalization (`/doc sync` for bidirectional normalization, `/doc add`/`/doc project` for adding documents, `/doc translate {lang}` for generating multi-language translations).

Issue creation and refinement (`/issue "title"` or `/issue 123`). Creates new issues or refines/reformats existing ones. Use when creating issues, defining requirements, or standardizing issue content.

1 hook across 1 event