Roberdan/roberdan-os

Copilot CLI-first workflows for AI agents: shared instructions, reusable skills, multi-model delegation, human approvals, and local-first task tracking. Fork your own.

Qu'est-ce que roberdan-os ?

roberdan-os is a Claude Code agent skill that copilot CLI-first workflows for AI agents: shared instructions, reusable skills, multi-model delegation, human approvals, and local-first task tracking. Fork your own.

Compatible avecClaude CodeCodex CLI~Cursor
npx skills add Roberdan/roberdan-os

Installed? Explore more Recherche et analyse de données skills: obra/superpowers, affaan-m/quarkus-verification, affaan-m/uspto-database · View all 6 →

Demander à votre IA préférée

Ouvre une nouvelle conversation avec cette compétence d'agent déjà préchargée.

Documentation

ship — get the work into production (git + gh)

Platform-independent ship workflow. No bypasses, no shortcuts.

Pre-push: local CI pipeline (everything green before pushing)

  1. Formatcargo fmt --check / prettier --check / ruff format --check
  2. Lintcargo clippy -- -D warnings / eslint / ruff check
  3. Type-checkcargo check / npx tsc --noEmit
  4. Testcargo test / npm test / pytest (output shown)
  5. Buildcargo build / npm run build

If a step fails: fix and re-run all checks. Never push with known failures.

Sequence

  1. Dedicated branch (never work directly on main).
  2. Commit per phase, conventional + evidence-first messages (SHA/PR/CI).
  3. git push the branch.
  4. Open PR: Summary + Test plan (5-section template if the repo uses one).
  5. Watch CI: gh pr checks <n> — all SUCCESS before proceeding.
  6. Merge: merge-commit only (never squash, never rebase — preserves history for parallel agents).
  7. Post-merge: delete the branch, fast-forward local main, report the merge commit SHA.

Human gates (STOP — ask first)

  • Merge to main impacting branch-protection / security / license / release-infra (#1)
  • Force-push to main (#2) — always forbidden without explicit confirmation
  • CI not fully green → don't merge anything pending/failing

Review comments

Every comment (human or bot) must be analyzed, understood, and resolved well — never silent-resolve, never a "fix" that only touches the quoted line while ignoring the substance. Reply on the thread with what you did and why, then resolve. See skills/review.

Skills associés