Write a bug report for an AI coding agent: reproduce, isolate and verify

A worked example: where the task becomes difficult

A fictional workspace crashes when a user switches from Build to Chat after selecting a file. Reloading temporarily restores it. The report should name the mode sequence, whether an edit was pending, the selected file type and the first relevant console error. Repeating “the editor broke” does not let the agent distinguish a render loop, stale callback or failed network request.

Decisions to make before implementation

Capture the earliest error associated with the failure rather than every unrelated browser message. Record app revision, browser version and whether the issue occurs in a fresh controlled project. Remove tokens, private file contents and complete personal identifiers from pasted logs. Preserve the meaning of URLs and record IDs with consistent placeholders so requests can still be followed across the report.

A practical sequence for the work

Use the sequence below as a task boundary, not as a claim that the example has been executed. Work with approved inputs and the project’s actual architecture. If a required integration or permission is unavailable, keep that stage visibly incomplete rather than generating a plausible substitute result.

• Write numbered reproduction steps starting from a known state.
• State the expected visible result and the first observed failure.
• Attach a minimal redacted trace with revision and browser context.
• Ask the agent to verify the same sequence after fixing the supported cause.

A detailed brief you can adapt for your agent

Replace the illustrative context with your approved facts and controlled inputs. Keep the stated boundaries when adapting the brief. The expected deliverable matters more than a particular tool name: ask for an explanation grounded in the inspected material and evidence for the requested outcome.

Investigate this mode-switch crash from the numbered sequence and trimmed trace. Find the component owning the first failure. Distinguish unrelated network warnings from causal evidence. Do not remove the error boundary or suppress errors as the fix. Reproduce the action after the change and report the result and revision.

Failure modes that an attractive preview can hide

A blocked advertising request may appear beside an application crash without causing it. Keep the evidence streams separate unless source or runtime behavior connects them. Avoid treating a generic error boundary as the root error. If the agent cannot reproduce with available files, request the missing context rather than accepting a patch that simply suppresses the error screen.

Technical references: GitHub: getting useful results from coding tasks

Acceptance checks and the evidence to retain

The report and resolution should include a reproducible sequence, supported cause, changed responsibility and observed retest. Keep unsupported explanations marked as hypotheses.

Specific answers

Common questions