Browser Probe

Use this skill for exploratory browser QA that stays lighter than a formal repro package.

Workflow

1. Pick a mode: smoke, balanced, or bug-hunt.
2. Create one named agent-browser session and reuse it for the whole run.
3. Model the app or scoped feature.
4. Generate likely failure hypotheses and likely strengths.
5. Execute focused checks with proportional evidence.
6. Report confirmed issues, verified strengths, flakes, and blocked areas.
7. Escalate to dogfood if the user now needs handoff-grade repro artifacts.

References

Read these in order:

• references/stage-1-model.md
• references/stage-2-hypothesize.md
• references/stage-3-execute.md
• references/stage-4-report.md

Use these only when needed:

• references/session-preflight.md
• references/recovery.md
• references/patterns/*

Guardrails

• Keep one live browser session per run.
• Re-snapshot after DOM-changing actions.
• Use proportional evidence, not artifact theater.
• Route repro-heavy handoff requests to dogfood.