Codex Debug: Bug Investigation

What this prompt does

Reproduce and isolate a reported defect with repository-aware safeguards. It is designed for bug investigation and works best with Current Codex coding model configured for the repository.

Customize and copy

This is a reusable natural-language prompt for Codex, not a shell command.

Work in the repository at [REPOSITORY_OR_PATH]. Goal: reproduce and isolate a reported defect. Context: [CONTEXT]. Constraints: [CONSTRAINTS]. Definition of done: [ACCEPTANCE_CRITERIA].

Reproduce first, gather evidence, rank hypotheses, and change code only after identifying the likely cause.

Read applicable AGENTS.md guidance and use Codex’s inspection, editing, and verification workflow; keep approval boundaries and the user’s working tree intact.

First read the repository instructions and relevant framework documentation. Preserve unrelated user changes. State assumptions, cite files and lines for findings, and ask only when a missing decision would materially change the result. Do not invent APIs, commands, flags, test results, or file contents. Deliver: reproduction, hypotheses, root cause, and minimal fix. Run the repository’s relevant formatting, lint, type, test, and build checks when authorized, and report exact results.

How to use it

Replace the bracketed variables with real context for bug investigation. Attach or paste source material when available, and verify consequential facts before acting.

Example input

Provide a real bug investigation brief with the audience, goal, constraints, preferred format, and any available source material filled into the prompt’s placeholders.

Sample output

Reproduce and isolate a reported defect with repository-aware safeguards. The finished response should apply that result to bug investigation, follow the requested structure, distinguish supplied facts from inference, and end with usable next actions.

Quality checklist

• The response directly addresses the bug investigation goal and intended audience.
• Every requested section or output format is present and materially complete.
• Unsupported facts are removed, sourced, or clearly marked for verification.
• The final result is specific enough to use without guessing what happens next.

Limitations and review notes

Repository-aware agents can still misunderstand project conventions. Keep changes scoped, inspect diffs, and validate security-sensitive behavior manually.

This page records an editorial structure and safety review, not a guarantee that every provider or model version will produce the same result. See the Promptle testing methodology.

Related prompts