Crucible — Habilidades de Desenvolvimento Sistemático
raddue/crucible
52 skills para pipeline completo: design, TDD, review, debug e quality gates
Ver no GitHubColeção de 52 habilidades para todo o ciclo de vida do desenvolvimento de software: TDD, revisão de código, debugging, testes adversariais e gates de qualidade. Funciona com Claude Code, Cursor, Cline e outras plataformas compatíveis com SKILL.md. Reduz tokens através de despacho em disco e recuperação de falhas entre fases.
Instalação
npx convez-ai raddue/crucible1 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
- Orquestrar pipelines de código end-to-end com recuperação de falhas
- Implementar TDD com testes adversariais (red-team automático)
- Executar revisão de código iterativa com stagnation detection
- Testar segurança com 6 agentes atacantes paralelos (siege)
Veredito

Ferramenta madura e bem-testada (+49 evals com deltas mensurados de +29/+31%) para times que precisam de qualidade industrial em código. Exige familiaridade com orchestração de agentes e workflows em disco, não é ponto de entrada.
README
A tradução para o português deste README ainda está na fila.
Crucible
A collection of agent skills for systematic software development. Works with Claude Code, Cursor, OpenAI Codex, Amp, Cline, and any platform that supports the SKILL.md format.
52 skills across the full development lifecycle — design, planning, TDD implementation, code review, debugging, adversarial testing, and quality gates. 13 core skills are eval-tested with measured A/B deltas (49 execution evals + 18 sequence evals, +29% / +31% average).
Originally forked from obra/superpowers, now independently maintained and significantly diverged. Pipeline checkpoint system, auto skill extraction, structured context compression, and trajectory capture inspired by NousResearch/hermes-agent.
Marketplace
| Platform | Status |
|---|---|
| Claude Code | Pending review |
| skills.sh | npx skills add raddue/crucible |
| Cursor / Codex / Amp / Cline | Compatible — same SKILL.md format |
What you get
- Iterative quality gates, not single-pass review —
quality-gateloops until clean (red-team → fix → fresh re-review), with weighted stagnation detection. Measured +68% delta (see eval results). - Full pipeline orchestration —
buildchains design → plan → execute → complete with shadow-git checkpoints, crash recovery, and structured compaction-state blocks. One command, idea to merged PR. - Adversarial testing at every level —
adversarial-testerwrites tests designed to break the implementation;inquisitorruns 5 parallel attack dimensions against the full feature diff;siegeruns 6 attacker-perspective security agents until zero Critical/High. - Token-efficient by design — orchestrators use disk-mediated dispatch: full subagent prompts go to
/tmp, only ~100-token pointers enter orchestrator context. A full build saves 73-131K tokens of fossilized context. - Crash-resilient and session-continuous —
replayresumes interrupted pipelines from the last phase boundary (10 minutes, not 90);recallqueries a searchable session activity index that survives compaction. - Forge-agnostic iterative code review —
temperruns fresh-eyes review loops on PRs from any platform (GitHub, GitLab, Bitbucket, self-hosted) or raw<base>..<head>ranges. Multi-round convergence with stagnation detection; optional external-model second opinion via MCP. Renamed from/code-reviewto avoid collision with Claude Code's built-in/review.
See docs/architecture.md for how the orchestrator skills compose, and docs/skills.md for the full catalog.
Install
Claude Code
git clone git@github.com:raddue/crucible.git ~/repos/crucible
ln -sf ~/repos/crucible/skills/* ~/.claude/skills/
mkdir -p ~/.claude/agents
ln -sf ~/repos/crucible/agents/* ~/.claude/agents/
Or, when available on the marketplace: claude plugin install raddue/crucible. Note: the per-role model pins are currently delivered by the symlink install above — under a plugin install the agent types are namespaced (crucible:…) and bare-name dispatch resolution is unconfirmed (see harness-adapter §8).
Cursor / OpenAI Codex / Amp / Cline
git clone git@github.com:raddue/crucible.git ~/repos/crucible
Then point your platform at the cloned skills directory. See Cursor plugin docs and Codex skills docs. Some advanced features (parallel subagent dispatch, agent teams, persistent memory) are platform-specific and degrade gracefully — see PLATFORMS.md.
Setup essentials
Run in auto mode for autonomous pipelines (/build, /debugging, long /quality-gate runs). Auto mode lets Claude execute tool calls without per-action prompts while still confirming destructive operations. It supersedes earlier --dangerously-skip-permissions guidance.
Set CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 for build's parallel team execution. Without it, independent tasks fall back to sequential dispatch — everything still works, just slower.
Recommended but optional: enable the forge (retrospectives) and cartographer (codebase map) knowledge accelerators — they compound skill quality over time. See docs/architecture.md.
For hooks (session activity index, build routing advisor), MCP-based external model review, autocompact tuning, and other Claude Code specifics, see docs/configuration.md.
Documentation
- docs/skills.md — full catalog of all 52 skills
- docs/architecture.md — how the orchestrator skills compose
- docs/configuration.md — Claude Code settings, hooks, MCP, env vars
- docs/evals.md — full eval methodology and per-skill A/B deltas
- PLATFORMS.md — cross-platform compatibility notes