with one click
debug
Interactive debugging — reproduce, isolate, trace, fix
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Menu
Interactive debugging — reproduce, isolate, trace, fix
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Guided git commit workflow — review changes, draft message, stage, commit
Manage context pressure during large codebase work without losing critical state.
Create a pull request with structured summary and test plan
Explore a codebase or directory to understand its structure and purpose
Implement a requested code change end-to-end with scoped edits and verification.
Investigate production-like failures from symptoms, logs, and reproduction steps.
| name | debug |
| description | Interactive debugging — reproduce, isolate, trace, fix |
| when_to_use | When something doesn't work and the cause isn't obvious from reading code |
| required_tools | ["bash","symbol_search"] |
| tags | ["debugging","troubleshooting"] |
| activation | {"input_patterns":["(?i)(debug|doesn.*work|not working|怎么.*不|排查|调试)"]} |
${ARGS}
Run the failing scenario. Capture the exact error: message, stack trace, exit code, log output. If it cannot be reproduced, clarify the conditions under which it fails.
Narrow the scope. Does it fail with minimal input? Which component produces the first incorrect output? What changed between "it worked" and "it broke"? (git log, git diff)
Follow the actual execution path, not the assumed one. Add targeted logging at decision points. Read the code the runtime actually executes — follow imports, not mental models. Check configuration, environment variables, file permissions, network state.
Name the exact line or condition. Explain the mechanism: not "it's broken here" but "this value is null because the initialization path skips it when X is true."
Change only what the root cause requires. Run the original failing scenario to confirm. Run the full test suite for regressions. Search for the same pattern elsewhere in the codebase.