agentleFS
moronireismoronireis/agent-system-partners/.claude/skills/pipeline-sistema/SKILL.md

pipeline-sistema

Skill0 starsChanged 2 months agoMITView on GitHub

Full system-build pipeline: plan → spec → implement → QA gate → deploy

The file

Files
SKILL.md

Pipeline — Sistema

Runs the dev squad end to end: turns an objective into a deployed, tested system. Invoked directly as /pipeline-sistema, or picked up by director when a plan calls for full-system implementation rather than a single edit.

When to use

  • The objective needs new architecture — a stack, schema, or API decision that doesn't exist yet.
  • The work spans 2+ dev squad agents (e.g., a new table, plus new UI, plus a webhook).
  • The result will be deployed and needs a functional gate before it ships.

When NOT to use

  • Trivial single-file fixes (a typo, one CSS tweak, a copy edit) — edit directly, no pipeline overhead.
  • Pure design or content work with no code involved — use /pipeline-brand, /pipeline-content, or /pipeline-copy instead.
  • A landing page from an existing offer/copy brief with no new architecture — /pipeline-landing-page covers that chain directly and already ends in dev-builder → dev-qa → dev-ops.
  • The architecture is already specced and unchanged — skip straight to the implementation stage instead of re-running the full chain from director.

Chain

  1. director — plans the objective in the single-round format (Planning Doctrine, ~/.claude/CLAUDE.md): everything executable now ordered by dependency, the barrier, done criteria, usage estimate with routing options, plus one task contract per stage (docs/task-contract.md) → brain/ops/plans/<slug>.md. No calendar, no durations. Execution starts only on the user's GO. Never implements.
  2. dev-architect — technical design: stack, data schema, API contracts, module boundaries. Writes brain/ops/handoffs/spec-<slug>.md.
  3. Parallel implementation — each agent reads the spec independently and works only inside its own module boundary:
    • dev-data — schema, migrations, RLS → brain/ops/handoffs/data-model-<slug>.md
    • dev-builder — UI implementation from spec + design handoff → brain/ops/handoffs/build-<slug>.md
    • dev-integrations — external APIs, webhooks, automation plumbing → brain/ops/handoffs/integrations-<slug>.md
  4. dev-qa — gate. Runs the golden path, edge cases, and regressions against the combined output. Verdict PASS/BLOCK written to brain/ops/reports/qa-<slug>.md.
    • BLOCK routes the failure list to the named owning agent (dev-data / dev-builder / dev-integrations) for a fix, then re-tests. Max 2 revision loops; a third failure escalates to the user with the blocker stated plainly — nobody retries silently.
  5. dev-ops — deploy. Reads the dev-qa verdict first: BLOCK means no deploy. Preview deploys proceed on PASS. Production deploys additionally require explicit human approval, regardless of verdict.
  6. Report — the main session closes the round with the doctrine's 5-block report (~/.claude/templates/report-rodada.md) at brain/ops/reports/round-<slug>.md: done vs planned with evidence, test script for the human/client, requests for the next planning round, actual vs estimated usage, next barrier.

Handoffs

Every stage passes state forward as a file in brain/ops/handoffs/ — never as implied memory or a direct ask to the previous agent. Gate verdicts and deploy results land in brain/ops/reports/. Any agent, or a fresh session, can resume the pipeline mid-way by reading the latest handoff file instead of replaying the conversation.

Done when

brain/ops/reports/qa-<slug>.md shows PASS and brain/ops/reports/deploy-<slug>.md shows a successful deploy — or, for production, a successful preview plus a clearly stated pending approval — and brain/ops/reports/round-<slug>.md exists with the test script and the requests for the next round.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.