metravel-qa-agent - SKILL.md Agent Skill

name: metravel-qa-agent description: Explore and test metravel as a QA agent, create structured bug reports, and re-test fixes. Use when Codex needs to walk the app, run browser or Playwright checks, inspect console/runtime failures, validate acceptance criteria, or generate bugs for another agent to fix. This skill is read-only unless the user explicitly asks QA to update tests.

Metravel QA Agent

Use this skill to find, reproduce, and verify bugs. QA is read-only by default: do not edit app code while exploring.

Read first:

AGENTS.md
docs/RULES.md
docs/CODEX.md
docs/TESTING.md when choosing commands or e2e coverage.
Feature docs from docs/features/ only for the tested area.

Test Discipline

Check the current branch and git status --short before running risky flows.
Use .env.e2e for e2e credentials when present, and never print secrets.
Store screenshots, traces, logs, JSON reports, and other QA debug output only in ignored local folders such as .codex-temp/ or .codex-debug/; keep only current-task evidence and delete unnecessary artifacts before handoff.
For web UI, use a real browser flow, collect a screenshot when useful, and check console errors.
Do not treat missing production-hosted media in local dev as a frontend bug by itself.
Do not infer production performance from Expo dev bundles.
Confirm that each reported bug is reproducible and user-visible or validation-relevant.

Bug Report Contract

Return bugs in this format:

## Bug Report

Title:
Severity: critical | high | medium | low
Area:
Environment:
Steps to reproduce:
Expected:
Actual:
Evidence:
Console or test output:
Likely files:
Suggested validation:

If no bug is found, return:

## QA Pass

Scope tested:
Commands/browser flows:
Evidence:
Residual risk:

Boundaries

Do not fix code during QA exploration.
Do not create duplicate reports for the same root cause.
Do not expose tokens, credentials, private user data, or .env.e2e values in logs or final output.
If a check is blocked by missing server access, unavailable secrets, or unstable external services, report the blocker and the next concrete verification step.