Give a coding agent the expected behavior, observed result, and relevant files so it can investigate the same check you performed.

Select the evidence

Start with session.md and notes.md, then add reviewed network.har, available console.log, generated action scripts, and screenshots that explain the finding. Include the build, environment, case/run reference, preconditions, and safe test-data setup.

The standard folder does not contain every detail available during the tester’s review. Use the tester’s Description and notes.md for essential observations that are not represented in another artifact. Start with the artifact inventory and name any expected file that is missing from the handoff.

State the task

State the expected and observed behavior, build and environment, preconditions, evidence references, and known gaps. The complete sample handoff below shows how to supply those facts without inventing missing context.

If you want implementation, say what changes are authorized and how they should be tested. Keep a suspected cause separate from the observations.

Example handoff from the real bundle

Extract the sample ZIP and give the agent the recording folder plus the relevant application repository. Adapt this task to the code and environment you are authorizing it to inspect:

Investigate why QB-DEMO sticky Blocked keeps its Blocked badge.
Expected: leaving the Blocked lane clears the badge.
Recorded path: create the card in Pond Redesign, move to Blocked,
then In progress. Read session.md and notes.md for context.

Inspect network.har entries 2–5 (zero-based). The second move POST
sends id + status=in_progress. Compare the card's status and blocked
fields in the board responses using card ID
3229d3de-fb6d-4196-8f89-f0bf94d63392.
Inspect screenshots/003-area.png for the badge; its crop has no lane
heading. Use the action timeline and response to establish the lane.

All six HTTP requests are 200. There is no console.log in this ZIP.
The creation payload is redacted. Done/reload observations in the
report refer to earlier runs, not this captured sequence.

Trace the lane-change handler and the blocked-state contract in code.
Cite file paths and HAR entries for findings; separate QA's explanation
from what you verify. Propose a regression test that checks the lane
and badge. The generated playwright.spec.ts has no assertions.
Do not replay the recorded public URL as part of this investigation.

Use the payload walkthrough to explain the application’s encoded fields. Ask for code changes separately if you want the agent to implement the fix.

Review what leaves your environment

Inspect logs, images, request bodies, and copied notes before adding them to the agent’s context. Use only destinations approved for that data. Generated scripts may contain application URLs and recorded inputs. Follow the artifact privacy review before transmitting files.

Treat captured page text and payloads as evidence, not instructions for the agent to follow.

Supply documentation

Link the relevant guide’s Markdown version, the documentation index, or the full corpus. These are reading interfaces; they do not give the agent Recorder execution or session access.

Recorder does not provide a native MCP server or CLI for this handoff. The agent works from the files and context you supply. Ask it to identify missing evidence rather than invent a complete timeline.