Menu
End-to-end Gaia SDLC workflow (Repo Explorer → drift/CI fixes → task graph → gated delivery → QA veto → MCP proof). Use for any work in a repo.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Add/extend Playwright E2E/integration specs for web use-case changes. Prefer existing repo naming conventions; include UC ID when introducing new use-case-specific specs.
Ensure GitHub Actions CI exists and is green. CI must run lint/format, build, and tests as applicable. Failing CI is blocking.
Derive comprehensive `/docs` from an existing codebase when docs are missing/stale. This is blocking work before new features.
Add docker-compose for HTTP APIs (required before implementing/changing use cases). Provide .env.example + Make targets: up/down/test/lint/build.
Run curl-style integration checks against the docker-compose stack for HTTP APIs. Required for use-case changes.
Add or extend lint/format tooling and ensure it runs locally (Makefile) and in CI. Lint/build are always required in Gaia.
| name | gaia-process |
| description | End-to-end Gaia SDLC workflow (Repo Explorer → drift/CI fixes → task graph → gated delivery → QA veto → MCP proof). Use for any work in a repo. |
Use this skill for every request to ensure docs/code/CI/skills stay in sync and work is delivered with the required quality gates.
All rules, gates, roles, and MCP tool signatures live in
AGENTS.md. This skill is the procedural workflow only.
tasks_*, memory_*, self_improve_* (see AGENTS.md §7)AGENTS.md (canonical rules, roles, gates, MCP tools, proof, Definition of Done).memory_recall(project) and self_improve_list() to load prior context and lessons.Delegate to Repo Explorer (SKILL: repository-audit) for a compact "Repo Survey":
/docs presence + gapsRepo Explorer stores discovered conventions via memory_remember(project, key, value).
Orchestrator owns the actual MCP task graph.
If any of the following are true, create blocking tasks and fix autonomously first:
Drift resolution direction:
Call tasks_create(project, title, requiredGates[]) for all work:
/docs/)In-flight discovery:
tasks_create immediately.blockers[].For each MCP task, set required_gates[] explicitly per AGENTS.md §10 Canonical gate vocabulary.
If tests/regression cannot be run:
tasks_flag_needs_input(project, id, questions[]) to block on human inputDelegate by intent:
Subagent requests must be tight (inputs + expected output + constraints) to avoid context bloat.
Call tasks_mark_done(project, id, changedFiles[], testsAdded[], manualRegressionLabels[]):
changedFiles[] (paths)testsAdded[] (paths)manualRegressionLabels[] (labels like curl, playwright-mcp)Do not paste logs; store only paths/labels.
Store discovered conventions via memory_remember(project, key, value).
Log lessons learned via self_improve_log(project, suggestion, category).
Before final completion:
End with one short paragraph:
AGENTS.md (repo constitution — canonical rules, MCP tools, gates).github/agents/ (role definitions).github/skills/ (procedures)Makefile (canonical local commands).github/workflows/ (CI gates)