setup-codebase-harness
/setup-codebase-harness
Ocenia i poprawia środowisko pracy agentów w repozytorium: dokumentację, lintery, jedno polecenie uruchomienia i bramkę testów end-to-end.
- Autor
- thetangstr ↗
- Repozytorium
- agentdash ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 14.10.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 1
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 14.10.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Ocenia i poprawia środowisko pracy agentów w repozytorium: dokumentację, lintery, jedno polecenie uruchomienia i bramkę testów end-to-end.
Wyzwalacze z opisu
Oryginalny opis (EN)
Assess and improve the agent harness for a repo so an agent can work it reliably: legible (map-not-manual docs + custom lints), executable (one-command dev stack, worktree-friendly), verifiable (e2e gate + an independent verify-before-ship loop), plus the loops/ knowledge substrate so runs compound. Use when onboarding a repo to agent-driven development — "set up the harness", "make this repo agent-ready", "harness this codebase".
Jak działa
- 0. Assess
- 1. Legible — the agent can reason about the repo
- 2. Executable — the agent can run & drive the app
- 3. Verifiable — the agent can prove it works
- 4. Shared brain — runs compound
- 5. Keep it coherent over time
- Order & what you leave behind
Podgląd SKILL.md
--- name: setup-codebase-harness description: > Assess and improve the agent harness for a repo so an agent can work it reliably: legible (map-not-manual docs + custom lints), executable (one-command dev stack, worktree-friendly), verifiable (e2e gate + an independent verify-before-ship loop), plus the loops/ knowledge substrate so runs compound. Use when onboarding a repo to agent-driven development — "set up the harness", "make this repo agent-ready", "harness this codebase". user_invocable: true --- # Set up the codebase harness **Harness engineering:** the model is fixed — what you engineer is the *scaffolding* around it (the environment, the docs, the feedback loops, the shared memory) so an agent can build and verify software with minimal human attention. Humans steer; agents execute. Make the repo **legible, executable, verifiable** — and give it a **shared brain** so work compounds. Work **incrementally and depth-first**: assess what exists, build the one missing capability, use it to unlock the next. Don't boil the ocean. When the agent struggles, the fix is almost never "try harder" — ask *"what capability is missing, and how do I make it legible and enforceable?"* For AgentDash specifically, much of this already exists — this skill is mostly an **assessment + gap-fill**, not a from-scratch build. Current state (2026-06): | Pillar | AgentDash today | |---|---| | Legible | `AGENTS.md`/`CLAUDE.md` index + `doc/` specs ✅ · `scripts/check-architecture.mjs` lints ✅ | | Executable | `pnpm dev` (one command, :3100) ✅ · git-worktree workflow (`/workon`) ✅ | | Verifiable | `tests/e2e` Playwright gate ✅ · `/tester` independent verifier gate ✅ | | Shared brain | `loops/` substrate (signals/docs/domains + LOG) ✅ | ## 0. Assess Survey the repo: stack, package manager, services/ports, infra deps, existing docs/tests/CI, and the *implicit* rules (buried in READMEs, PR comments, people's heads). Note what's missing per pillar below before building anything. ## 1. Legible — the agent can reason about the repo > What the agent can't see doesn't exist. Knowledge in chat threads / heads is invisible — push it > into versioned, repo-local artifacts. - **a) Map, not manual.** Keep the root agent doc (`AGENTS.md` / `CLAUDE.md`) a slim table of contents (overview, tree, golden rules, a "where to look" table). Depth lives in `doc/`. AgentDash caps `CLAUDE.md` at 200 lines by policy — honor it. - **b) Custom lints with remediation.** Promote prose golden rules into **mechanical checks** whose error message **injects the fix**. AgentDash uses standalone `scripts/check-*.mjs` (e.g. `check:architecture`, `check:tokens`) wired into `.github/workflows/pr.yml` — add a new rule by pushing onto `RULES` in `scripts/check-architecture.mjs`, with a test in the sibling `.test.mjs`. One lint per invariant. ## 2. Executable — the agent can run & drive the app - One-command stack: `pnpm dev` (server + UI, :3100). State reset: `rm -rf data/pglite && pnpm dev`. - **Worktree-friendly** so parallel agents don't collide — the MAW `/workon` flow creates a per-issue worktree and warms deps. Carry over gitignored `.env` files into new worktrees. - Make the app drivable: Playwright / Chrome automation; logs reachable. ## 3. Verifiable — the agent can prove it works - **e2e gate** (`tests/e2e`): real flows over bypass, a reusable auth/session helper, layered client→server→product assertions, video/trace evidence, sandbox-only external services. Critical flows that must never break: signup, upgrade/billing. (Billing e2e is a known gap — add it.) - **Verify-before-ship**: an **independent** read-only verifier drives the running app to confirm the feature works — never self-verify. In AgentDash that's the `/tester` gate (Phase 4 CUJ + video proof). The verifier judges; the builder fixes until green. - **Proof, not claims**: attach the e2e/CUJ video to the PR via the `pr-evidence` gh prerelease so a human can watch the feature work. ## 4. Shared brain — runs compound - `loops/` is the substrate: `signals/` (evidence), `docs/` (durable knowledge), `domains/` (loops), `LOG.md` (global feed). Every loop reads the last few LOG entries before starting and appends one when it finishes; friction/ideas get filed as signals. See `loops/ARCHITECTURE.md`. - Stand up a new recurring loop with the `new-loop` skill. ## 5. Keep it coherent over time - **Commit hygiene** + light merge gates (at high agent throughput, corrections are cheap, waiting is expensive). **Agent-to-agent review** for correctness-critical changes (independent reviewers, not self-review). Periodic small refactor/cleanup PRs rather than debt in painful bursts. ## Order & what you leave behind Assess (0) → fill the missing pillar. For a fresh repo: **map → dev-up → e2e + verifier → lints → shared brain.** The artifacts — slim map + `doc/`, `pnpm dev`, an `e2e/` suite, the verifier gate, `check-*.mjs` lints, and `loops/` — are each a reusable capability that compounds. Prefer boring, composable, stable tech the agent can fully model.
Wersje
W bazie nie ma innych skilli o nazwie setup-codebase-harness.
Migawka bazy z 13.11.2026 · wersja paczki 8f6355489f04 (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 100harness-factory
Sprawdza gotowość repozytorium do pracy agentów, a potem buduje jego harness i fabrykę oprogramowania, krok po kroku, z weryfikacją każdego etapu. - 100harness-repo
Przygotowuje repozytorium do pracy agentów: krótki AGENTS.md, baza wiedzy docs/, linter zasad, CI dla stosu i plan wykonania przez skrypt. - 100harness-engineering
Zakłada lub aktualizuje uprząż inżynierską zorientowaną na agentów w repozytorium: mapa wiedzy w AGENTS.md, ustrukturyzowana dokumentacja, granice architektury, reguły .harness/*.yml i ocena jakości. - 90harness-setup
Inicjalizuje albo aktualizuje konfigurację Oh My Harness w projekcie Codex. - 100review-agent-harness
Ocenia, czy harness agenta kodującego prowadzi pracę od intencji przez wykonanie, weryfikację, dostarczenie i naukę, i gdzie zawodzi. - 90coding-agent-harness
Projektuje harness agenta kodującego: ograniczenia, weryfikację i bazę akceptacji oraz ocenia, które zadania nadają się dla agenta.