See the causal chain that led to an error
I don't just want the error — I want the story: what I clicked, what the DOM did, and which request failed, in order, right before it broke.
What you’ll end up with
For a chosen error, a single time-ordered timeline — your actions, the DOM mutations, and the network errors in the window before it — plus a short deterministic narrative of how the error was triggered.
Prerequisites
- Claude Code or Cursor with peek wired in (
peek initadds the MCP entry to~/.claude.json) - Chrome with the peek extension installed — from the Chrome Web Store
- A recorded session that contains at least one console error
See Set up peek with Claude Code rather than restating the config here.
Steps
1. Pick the error
List the console errors in my latest peek session.
The agent calls get_session_console_errors and you (or it) pick the row id that matters.
2. Ask for the chain
For that error, show me the causal chain — what I did, what changed on the page, and what failed, right before it.
The agent passes the id to get_user_action_before_error. It returns one merged timeline (plus grouped actions, domMutations, and networkErrors, the seed error, and a narrative) for a window before the error.
3. Widen or narrow the window
Widen the window to 10 seconds before the error.
The agent re-calls with a larger windowMs to pull in earlier context.
Why this works
peek doesn’t just store the error — it reassembles the events around it from the recorded session, on-device. The chain is pre-merged and ordered, so the agent reasons over “action → DOM change → failed request → error” instead of stitching three separate queries together.
Good to know
The chain is bounded for signal: it keeps up to the most relevant DOM mutations and network errors in the window and flags truncated when there were more — so treat it as the story around the error, not an exhaustive log.
Next steps
Once you understand the trigger, capture it as a test with Generate a Playwright repro from a real session.
Trust & data handling
Read-tier (Level 1): the chain is assembled from your recorded session in ~/.peek, non-mutating. Local-first: peek uploads nothing — what your MCP client does with the data is up to you. Captured values are masked at record time.