metravel-codex-orchestrator

name: metravel-codex-orchestrator description: "Apply the metravel Codex operating system before and during work: task triage, minimal skill selection, role prompt selection, docs/rules checklist, validation planning, handoff format, and final self-check. Use when Codex needs to organize its own work, choose agents/skills, translate a user request into a safe workflow, or keep complex metravel tasks aligned with AGENTS.md and docs/CODEX.md."

Metravel Codex Orchestrator

Use this skill as the top-level Codex work controller for metravel tasks. It does not replace role skills; it chooses and sequences them.

Read first:

• AGENTS.md
• docs/CODEX.md
• docs/RULES.md
• docs/README.md
• Extra docs only when the triage table in docs/CODEX.md says they matter.

Start Prompt

At task start, silently or briefly answer:

Task type:
User-facing scope:
Required skills:
Docs to read:
Branch/status:
Risk zones:
Validation plan:
Operation gate:
Needs browser/mobile/native check:
External links affected:
Can proceed without clarification:

If the branch is not main, stop before editing and ask how to proceed.

Skill Selection Rules

• Always choose the smallest useful skill set.
• Keep docs-only changes, simple automation, narrow checks, and isolated bugfixes in one specialist skill unless the scope is unclear or high-risk enough to need orchestration.
• Use $metravel-project-analyst for broad or unclear repository analysis.
• Use $metravel-business-analyst before implementation when the request is a product idea or requirements are vague.
• Use $metravel-system-architect when design, module boundaries, risk mapping, or high-risk review is needed.
• Use $metravel-domain-router before implementing domain-heavy travel, map, profile, achievements, quests, PDF/export, new-page, or design-system work.
• Use $metravel-feature-builder for feature, bugfix, refactor, API, service, hook, SEO, or ordinary code changes.
• Use $metravel-hook-builder when the main work is hook extraction or hook-boundary cleanup.
• Use $metravel-refactor-surgeon for behavior-preserving extraction of large components or file-complexity guard failures.
• Use $metravel-ui-guardrails for visible UI, layout, icons, images, placeholders, tokens, or external-link interaction.
• Use $metravel-browser-reviewer for visible web diff review, browser evidence, console/network checks, and fix/reverify loops.
• Use $metravel-android-developer for Android/native implementation, crashes, Expo modules, permissions, SecureStore, push, or platform files.
• Use $metravel-mobile-tester for read-only mobile web or Android/native QA evidence and retest.
• Use $metravel-test-runner to choose and run targeted checks; $metravel-test-writer to add stable tests.
• Use $metravel-e2e-runner for Playwright/browser flows and .env.e2e auth.
• Use $metravel-performance-analyst only with production build or real URL performance evidence.
• Use $metravel-seo-index-operator for GSC/indexing routines, URL Inspection/index status, IndexNow backup, and SEO owner/action lists.
• Use $metravel-release-checks and $metravel-devops-agent for release/deploy/build operations and explicit environment gates.
• Use $metravel-production-smoke for read-only production health checks and route failures to the right owner.
• Use $metravel-docs-maintainer for docs, AGENTS, CODEX, or skill changes.
• Use $metravel-backend-diagnostician for read-only backend/API diagnosis and backend board follow-up.
• Use $metravel-sprint-reviewer for task-board acceptance and Done-gate verification.
• Use $metravel-code-reviewer before handoff when review findings, residual risk, or rule compliance matter.
• Use $metravel-quality-fixer when the task is explicitly to run/fix the full quality gate.

Role Prompt Pattern

When handing work to a role skill, keep the prompt compact:

Use $<skill-name> for <scope>.
Context: <relevant files/docs already known>.
Constraints: follow AGENTS.md, docs/RULES.md, docs/CODEX.md; keep unrelated changes separate; do not print secrets.
Output: <role artifact name>.
Validation: <expected checks/evidence>.

Do not leak desired conclusions into QA/reviewer prompts; pass raw scope, diff, logs, or evidence.

Work Rules

• Prefer existing components, hooks, services, utilities, and tests before adding abstractions.
• Keep edits small and local.
• Fix real issues found in the touched scope before handoff.
• Keep temporary screenshots, logs, traces, and JSON output only in ignored folders such as .codex-temp/, .codex-debug/, test-results/, or playwright-report/.
• Never print secrets from .env*, .env.e2e, EAS, SSH, Google Play, or server configs.
• Do not create one-off reports unless the user asks; update canonical docs instead.
• For visible web UI, require browser preview, screenshot, and console check.
• For Android/native, do not claim device readiness without emulator/device evidence.
• For production deploy or submit, require explicit target environment and clean gates.
• Before assigning or running deploy, build, server rebuild/restart, full/preflight tests, Playwright/e2e, or Lighthouse work, apply the operation coordination rule from AGENTS.md/docs/RULES.md; do not launch duplicates for the same target.

Final Self-Check

Before final response, verify:

• Scope stayed within the user request.
• Current branch rule was respected.
• Changed files are intentional.
• Relevant checks ran, or blockers are explicit.
• External-link, UI, native, release, and secret rules were not violated.
• Any known real issue in touched scope is fixed or documented as blocked.
• Final answer lists changed files, validation, and residual risks without dumping secrets or noisy logs.