Vigil Skills
ziomancer/vigil-skills
Workflow spec → review → ship com revisores paralelos para Claude Code
Ver no GitHubConjunto de skills e subagentes para Claude Code que implementa um workflow completo de especificação, revisão paralela e entrega de código. Automatiza ciclos de spec com 3 lentes de revisão (correção, casos extremos, convenções), implementação, testes e PRs, integrado com Plane.so e CodeRabbit.
Instalação
npx convez-ai ziomancer/vigil-skills1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Automatizar ciclos de especificação com revisão multi-ângulo antes de implementar
- Gerenciar implementação de specs, testes e criação de PRs de forma isolada
- Processar feedback de CodeRabbit e resolver issues de forma automatizada
Veredito

Ferramenta robusta e opinativa para equipes que querem rigor no workflow de desenvolvimento com Claude. Exige integração com Plane.so, GitHub e entendimento de git worktrees — não é plug-and-play, mas oferece controle real sobre qualidade de specs e entrega.
README
A tradução para o português deste README ainda está na fila.
Vigil Skills
Cross-machine skills and subagents for Claude Code. A Plane.so-aware spec → review → ship workflow with parallel reviewers, plus a CodeRabbit triage handler.
What's here
Skills
/spec-cycle <brief-path>— Author a spec from a brief, then run a 3-lens parallel review loop (correctness / edge-cases / repo-conventions) until findings clean or 4 passes complete. Halts at a session boundary with a structural drift-check checklist before any implementation. Pair with/ship-spec./ship-spec <spec-path>— Take a green-lit spec through implementation, test gate, PR, and Plane update. Cuts an isolated git worktree from your default branch (your primary working tree is never touched), implements + tests in a tight loop, captures test output for the PR audit trail, and pushes a PR./review-pr [<num>]— Process one round of CodeRabbit review findings on a GitHub PR. Triages by severity, fixes real issues, pushes, resolves threads, and verifies the resolve actually took.
Subagents
The spec-cycle skill dispatches three reviewers in parallel:
spec-reviewer-correctness— Does the spec actually solve the brief's problem? Verifies every claim about current code by reading the actual files.spec-reviewer-edge-cases— Empty/null/zero-length inputs, concurrency, external-system failures, runtime preconditions, observability/tripwire requirements.spec-reviewer-conventions— Repo conventions (CLAUDE.md), prior decisions (your wiki, if any), premature abstractions, reuse vs. duplicate, unneeded backwards-compat.
Install
git clone https://github.com/<your-username>/vigil-skills.git
cd vigil-skills
python sync.py install
This copies repo content into ~/.claude/skills/ and ~/.claude/agents/. Files in those directories that aren't in this repo (e.g., third-party skills installed separately) are preserved by default.
Sync workflow
| Command | Direction | Use case |
|---|---|---|
python sync.py install (or pull) | repo → ~/.claude/ | Daily driver after git pull. |
python sync.py push | ~/.claude/ → repo | When you've edited a skill in place and want to commit back. Only touches files already tracked in the repo. |
python sync.py status | (none, diff only) | Show what differs between repo and ~/.claude/. |
Flags: --dry-run, --verbose, --prune (install only — deletes files in ~/.claude/ not in repo, off by default), --claude-dir <path> (overrides $CLAUDE_CONFIG_DIR / ~/.claude).
Customizing for your project
These skills delegate project-specific rules (test commands, await guardrails, lint conventions) to your project's CLAUDE.md. The skills read it during preflight and respect what's there. See docs/customizing.md for details.
Cross-harness portability
Skills are authored once as Claude Code SKILL.md and meant to run on any harness in the fleet. docs/portability-contract.md defines what makes a skill portable — the canonical source format, the portable frontmatter subset, the requires: capability declaration, the intent-over-implementation rule, and the behavioral-parity definition. It is the contract every cross-harness item targets.
docs/authoring-portable-skills.md turns that contract into day-to-day authoring discipline, and lint.py mechanically guards it (python lint.py warn-only; python lint.py --strict to gate). Run it before committing a new or changed skill.
The engine that emits these canonical SKILL.md sources as per-harness packages is owned and kept in a separate repo — vigil-converter (Bun/TypeScript, deliberately outside this stdlib-only tree). It is the converter every per-harness adapter builds on (VHS-19).
Requirements
- Python 3.8+ for
sync.py(stdlib only — no pip install). - Claude Code — skills and subagents are Claude Code features.
- For
/spec-cycleand/ship-spec: a Plane.so workspace with the plane-proxy MCP server, andghCLI authenticated. - For
/review-pr:ghCLI authenticated, CodeRabbit configured on your repo.
License
MIT — see LICENSE.