Skip to content

Debugging and code review.

Start with reproducible evidence and request a patch you can inspect and verify.

You will prepare: Reproduction steps, patch scope and verification requirements

Open this brief in workspace →Editorial guide updated

Prepare these inputs

These are editable example inputs, not completed results.

Problem and expected behavior
Form submission intermittently returns 500; expected a success or explicit error
Available evidence
I will supply steps, errors and relevant code; identify missing evidence first
Permitted scope
Only the form and its route; no identity, billing or dependency changes
Verification
Reproduce the issue, add a relevant regression test, typecheck and review the diff

Work through the steps

  1. 01

    Describe a reproducible difference

    List input, actions, expected and actual behavior, timestamps and necessary logs without usable secrets.

  2. 02

    Check tool access

    A web chat may not read a repository or run commands. Establish actual file and execution access.

  3. 03

    Save the implementation brief

    Organize problem, evidence, scope and verification in a saved workspace version.

  4. 04

    Keep the patch tied to the cause

    Verify the hypothesis and patch the smallest relevant area. Surface unresolved scope decisions rather than rewriting adjacent modules.

  5. 05

    Review actual evidence

    Read the diff, test output and exit codes. Distinguish fixed, deployed and unverified states in the handoff.

Reusable prompt brief

Investigate {{problem}}. Available evidence: {{evidence}}. Permitted edit scope: {{scope}}. Verification requirements: {{verification}}. Inventory accessible files and reproduction steps without assuming a root cause. Propose the smallest fix and preserve unrelated uncommitted work. Do not change auth, sessions, billing, credits or dependencies without explicit authorization. If editing and execution are available, report the diff, commands and exit codes. Otherwise provide a reviewable patch and commands to run. Never report unrun tests as passing.
Fill the fields and save a version →