--- name: harness-help description: Router over the harness skills — which skill fits your situation, and the main flow through them. NOT for performing the routed work itself — it only points; invoke the skill it names. disable-model-invocation: true tools: Read --- # /harness-help You don't remember every skill, so ask. Most work travels one **main flow**; everything else is standalone. Each skill below is documented in its own `skills//SKILL.md` (core) or `extras/skills//SKILL.md` — this router only says *when to reach for which*. ## The main flow: idea → shipped 1. **`/spec `** — turn a fuzzy request into `spec.md` + `plan.md` under `.agent/plans//`, then route to plan approval (ExitPlanMode). Add `--interview` when the request is fuzzy enough that a wrong guess would commit the spec to the wrong shape, or `--score-candidates` when the field of problems/approaches is wide enough that the pick should be scored, not asserted. Approval is what unlocks editing — `spec-gate` stops asking once the plan-approval flag exists. 2. **`/supervise `** — run the approved plan wave by wave, dispatching specialists per the delegation contract and auditing after each wave. `--goal-mode` for budgeted, resumable runs. 3. **`/verify-completion `** — before accepting any "done": a fresh context re-checks the claim mechanically, then a refute-by-default judge. `/supervise` calls this at its completion step; reach for it directly whenever a builder's word is the only evidence. 4. **`/wrap`** — commit + PR with the safety gates (gitleaks, risk areas) intact. `--auto-push` / `--auto-merge` are opt-in escalations. A single small edit needs none of this — just make the edit; the gates scope trivial work out by design. ## Standalone - **`/council-review [--staged|--head|]`** — multi-vendor second opinion on a diff (Claude reviewer plus codex/gemini lanes), synthesized with citation checks. - **`/project-init`** (command) — scaffold a consumer project with `CLAUDE.md`, rules, and `gitleaks.toml`. - **Security** — diff-level security review is the `security-reviewer` agent (auto-dispatched on risk paths); a whole-repository or branch vulnerability scan is the official `/claude-security` plugin when installed (`/plugin install claude-security@claude-plugins-official`). ## Extras plugin (`agent-harness-extras`, optional) Rarely used skills live in a second plugin so the core stays light. Install once with `/plugin install agent-harness-extras@agent`; its skills appear namespaced as `/agent-harness-extras:` (the short `/` works when unambiguous — note `loop` collides with Claude Code's built-in `/loop`, so use the namespaced form). Each is documented in its own `extras/skills//SKILL.md`: - **`/harness-audit`** — read-only health check of the harness itself: one `verify-all.sh` dry-run, a per-check table, the doc-reality verdict named. - **`/manager-audit `** — meta-audit of a `/supervise` run (restatement, routing waste, token spend, role compliance). - **`/persona-review `** — a panel of grounded Korean citizen personas reacts to a piece of UX/copy/content; a user-perspective lens, not a code review. - **`/reorg-sync `** — after a directory move, sweep orphaned absolute-path references; dry-run first. - **`/code-tour [path|range]`** — write a human reading path through a codebase or a just-finished build; edits no source. - **`/worker-setup`** — install → auth → verify onboarding for the cross-vendor worker lanes (codex, antigravity), with a cost briefing before anything installs. - **`/loop `** and **`/harness-loop`** — bounded-attempt improvement loop (generic, and specialized to this harness's reviewer prompts). - **`/brain-ingest`** — distill quarantined brain `raw/` captures into curated notes. ## The gates underneath Skills are the methodology; hooks are the enforcement. When one interrupts you mid-flow, it names its own escape: - **spec-gate** asks for a plan on a substantive edit → run `/spec`. - **tdd-guard** blocks new prod code with no failing test → write the test first. - **pre-tool-guard / risk-area mutexes** stop destructive or contested operations → the user drives those; don't route around them. ## Keeping this router honest Any skill added to or removed from `skills/` or `extras/skills/` updates this file in the same commit — a router that lies is worse than no router.