Skills Claude Code - Pipeline de Entrega Autônoma
username-workspace/skills
Pipeline de entrega autônoma com revisão e CI para Claude Code
Ver no GitHubColeção de skills reutilizáveis para Claude Code que implementa um pipeline de entrega autônoma com commit automático em marcos, revisão adversarial de merge requests, monitoramento de CI em background e validação de correções de bugs. Cada skill funciona independentemente, mas se compõem em um fluxo acoplado via estado git.
Instalação
npx convez-ai username-workspace/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 commits em marcos de desenvolvimento e abrir PRs quando realmente prontas
- Revisar merge requests com análise adversarial e feedback iterativo
- Monitorar CI de PRs em background e retornar veredictos à sessão
- Validar correções de bugs com provas (testes falhando antes, passando depois)
- Auditoria de segurança com Trivy e relatórios priorizados
- Rastrear métricas de produtividade e qualidade via histórico git
Veredito

Conjunto bem engenheirado de skills que resolvem um problema real: acelerar ciclos de commit, review e deploy sem sacrificar qualidade. Requer familiaridade com git, CI/CD e Claude Code para máximo valor. A composição em pipeline é elegante, mas o setup inicial é técnico.
README
A tradução para o português deste README ainda está na fila.
skills
Open, horizontal, re-usable Claude Code skills & plugins by Username (Studio A.I.). Company-specific (vertical) skills live in a separate private repo.
Install
/plugin marketplace add username-workspace/skills
/plugin install <plugin>@username
The delivery harness
Four of these plugins compose into an autonomous delivery pipeline — each one useful alone, loosely
coupled through .git state when together:
ship-when-done commits at each milestone and opens the draft PR only when the work is provably
done (quality gate green) → merge-review holds every push until an adversarial review passes →
mr-watchdog watches the PR's CI in the background and brings the verdict back into the session
(ok, all good — CI green) → proof-of-fix enforces evidence-first bug fixing: a recorded probe
that fails before the fix and passes after it.
Plugins
| Plugin | Version | Category | What it does |
|---|---|---|---|
claude-remote-spawn | 1.3.0 | Agents | Spawn a new Claude Code session remotely, on your own machine. |
find-session | 1.0.4 | Agents | Find and resume the past Claude Code session you're thinking of. |
delivery-metrics | 1.0.6 | Analytics | Turn git history into a developer productivity & quality dashboard. |
aws-remote-auth | 1.0.3 | DevOps | Re-authenticate to AWS from anywhere, on demand, with an autofill device code. |
mr-watchdog | 2.3.2 | DevOps | Open a merge request, then forget it — a background watcher follows its CI and hands the verdict back to your session. |
ship-when-done | 1.6.5 | DevOps | Commit at each milestone, push so nothing is lost, open the PR when it's actually done. |
coding-agent-usage | 1.2.1 | FinOps | See your AI coding-agent usage — and where you rank against other developers. |
merge-review | 1.3.4 | Quality | An adversarial reviewer that scores the diff, fixes what's attested, and loops until it's merge-ready. |
proof-of-fix | 1.0.3 | Quality | Prove the bug before fixing it — then prove the fix with the same probe. Red before, green after. |
security-audit | 1.3.3 | Security | One Trivy scan, every ecosystem — a prioritised security report. |
Quality
Every plugin ships a hermetic test suite — bash scripts/run-tests.sh runs them all (CI gate).
Real usage is validated separately: a generative E2E lane (tests/e2e/) replays full deliveries —
seeded scenarios, project archetypes, human-divergence twists — against a real sandbox forge, and
every proven situation is recorded in tests/e2e/coverage.json. Engineering rules (the incident
rule, symptom vs root cause) live in CLAUDE.md.
License
MIT — see LICENSE.