Agent Prompt Pack

Use Revyl MCP tools only.

Primary objective:
Immediately start a live dev loop, then authenticate/explore with screenshot-grounded actions.

Hard rules:
1) FIRST tool call must be `start_dev_loop`.
2) Do NOT call `list_tests`, `list_remote_tests`, `list_variables`, or `list_env_vars` unless I explicitly ask.
3) Never assume state. Before important actions, re-anchor with `screenshot`.
4) Use this strict loop:
   - screenshot
   - one-line description of what is actually visible
   - one best action
   - screenshot to verify result
5) Short deterministic burst is allowed (max 2 actions), then re-anchor with screenshot.
6) If the screen is not what you expected, stop and re-anchor instead of continuing the prior plan.
7) If `manual_step_required=true`, pause and ask me to complete the manual step using `deep_link_url`, then continue.
8) Never print raw credentials/secrets.

Inputs:
- platform: {ios|android}
- app_id: {optional}
- build_version_id: {optional}
- goal: <exact flow to complete>
- keep_session_open: {true|false}

Final output:
- final visible app state
- ordered actions taken
- issues/anomalies found
- pass/fail by milestone

Use Revyl MCP tools only.

Goal: <describe exact goal>.

1. Call start_dev_loop (set platform if needed).
2. Read output and store session_index.
3. If viewer_url is present, print it for human visibility.
4. If manual_step_required=true, pause and ask me to complete the manual step using deep_link_url, then continue.
5. Execute strict loop until goal is complete:
   - screenshot
   - one-line observation of what is visible
   - one best action
   - screenshot to verify
6. Never do blind action chains. Max 2 deterministic actions, then re-anchor with screenshot.
7. If blocked, diagnose likely root cause and suggest smallest fix.
8. After success, convert the flow into a reusable Revyl test:
   - test name (kebab-case)
   - preconditions
   - ordered steps
   - ordered validations
   - variables
9. Call stop_dev_loop.
10. Final output:
   - final app state
   - actions taken
   - bugs/anomalies found
   - proposed regression test cases

Use Revyl MCP tools only.
Goal: <describe goal>.
For each step:
1) take screenshot,
2) describe visible state in one short line,
3) take one best action,
4) take screenshot to verify.
Repeat until goal is complete.
At the end, summarize final state, actions taken, and issues found.

Use Revyl MCP tools only.
Goal: bypass login and reach the home screen.
Keep the loop strict: screenshot -> observe -> action -> screenshot verify.
If blocked, diagnose likely root cause and propose the smallest fix.

Use Revyl MCP tools only.
Goal: reproduce and isolate the bug on device.
For each step: screenshot, short observation, one action, screenshot verification.
When reproduced, provide:
1) exact repro steps,
2) expected vs actual behavior,
3) most likely root cause area,
4) suggested fix direction.

Pause actions. Re-anchor now:
1) screenshot,
2) describe current state,
3) pick one action with reason,
4) screenshot verify.
Then continue with the same loop.

Take the successful exploratory session and convert it into a production-ready Revyl test.
Output:
1) test name (kebab-case),
2) preconditions,
3) ordered instruction steps,
4) ordered validation steps,
5) variables required,
6) pass criteria.
Favor stable validations over overly specific transient details.