Inspect Elements

For human visual feedback, the development build includes Agentation. Activate its toolbar, select elements, add notes, and copy the resulting selectors and context into the task. Inspect the referenced code before changing it. The toolbar is suppressed when window.__PROFILING__ is set before navigation and is absent from production builds.

Map the requested live DOM node to source using the dev-only window.__ELEMENT_SOURCE__ helper. Use an existing compatible dev server in this worktree or record ownership of one started for the task; production does not expose the helper.

Reuse a compatible session supplied by the calling task when it already owns the browser slot; use its exact name and leave its lifecycle with that caller. Otherwise open an owned session through ./scripts/pw-session.sh. Use -s=<session> on every command. The examples use inspect; substitute the actual name and URL, and skip open when reusing a session:

./scripts/pw-session.sh open inspect https://5chan.localhost/#/all
playwright-cli -s=inspect eval "window.__ELEMENT_SOURCE__?.ready ?? false"
playwright-cli -s=inspect snapshot
playwright-cli -s=inspect eval "async el => JSON.stringify(await window.__ELEMENT_SOURCE__.resolve(el))" e7

Choose e7 from the current snapshot; do not reuse a stale ref. If the helper is not ready, check window.__ELEMENT_SOURCE__?.error and allow its dev import to complete. Report persistent errors or missing helpers instead of guessing an attribution.

The result's source, componentName, and stack identify the node's file/line and React ownership. Inspect the source before editing; an attributed file is a starting point, not proof that it causes the reported behavior. For a compact trace, use window.__ELEMENT_SOURCE__.formatStack(info.stack, 5) on a resolved result.

When a snapshot ref is impractical, use resolveBySelector(selector) for a precise selector or resolveAtPoint(x, y) for viewport coordinates. If source is null, inspect useful stack frames or a nearby parent. Report an unresolved node when both are empty.

Close the exact session in cleanup only when this inspection opened it, including on resolution failure. Leave a caller-owned session open for the caller to continue and close:

./scripts/pw-session.sh close inspect

For a new session, a slot owned by another task is a reason to wait or use the wrapper's bounded wait, never to bypass the lock. Do not wait on the calling task's own session; reuse it as described above. Stop only a server this task started. Use the playwright-cli skill for session operations and profile-browsing only when performance measurement is part of the request.