← ClaudeAtlas

airlock-verifylisted

Use inside AirLock to verify a feature end-to-end — drive it, then confirm the backend and renderer did not error and the UI rendered correctly.
RicardoRamosT/Airlock-IDE · ★ 0 · AI & Automation · score 56
Install: claude install-skill RicardoRamosT/Airlock-IDE
# Verifying an AirLock feature You are running inside AirLock and can drive + inspect it via MCP tools. Use this loop to verify a feature actually works — the UI can look fine while the backend throws. 1. **Baseline.** Record the current time as an ISO-8601 UTC timestamp (run `date -u +%FT%TZ` in a terminal). Everything after is "since" this. 2. **Drive the feature.** Exercise it with the driving tools: `open_tab`, `open_terminal` + `send_terminal_input`, `set_pref` to flip a feature on (e.g. `{ "key": "quotaMeter", "value": { "enabled": true } }`), `start_dev_server`, `run_command`, `open_app_page`. To drive/verify a SIDEBAR panel (Extensions, Secrets, Git, …), focus it with `set_pref { key:"sidebarVisible", value:true }` then `set_pref { key:"activeView", value:"extensions" }`, then `capture_screenshot`. 3. **Check the backend AND renderer.** Call `read_events` with `{ "level": "error", "since": "<baseline>" }`. ANY result is an error the UI may have hidden — report it with its `op` and message. Renderer crashes show up here too (category `renderer`). Use `get_terminal_tail` for command output. 4. **Check the frontend.** Call `capture_screenshot` and actually look at the rendered UI against what the feature should show. 5. **Report.** For each acceptance criterion: pass/fail, the evidence (event `op`s, screenshot observations, terminal output), and any errors found. Notes: - `capture_screenshot` and `set_pref` require Self-verification