Systematic debugging
Skill ZC0325/sol-native-guardrails/plugins/sol-native-guardrails/skills/systematic-debugging
Five narrowly scoped Codex engineering skills that preserve native model judgment.
npx -y skills add ZC0325/sol-native-guardrails --skill systematic-debuggingAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- 25 days oldThe repository was created 25 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Diagnose reproducible bugs, failing tests, performance regressions, integration failures, or unexplained behavior that requires evidence and root-cause tracing before a reliable fix. Use for nontrivial technical failures; use a direct fast path when the cause is already demonstrated and local.
SKILL.md
1.9 KB, 308 tokens by cl100k_base, as published. Nobody here has run it
Systematic Debugging
Use evidence to locate the cause before changing behavior. Scale the investigation to the uncertainty and blast radius.
Fast Path
Use a direct fix when the failure is local, reproducible, and the cause is demonstrated by an error, test, diff, or minimal inspection. Make the smallest relevant change and run targeted verification.
Investigation Path
- Reproduce the symptom or collect the best available evidence.
- Read errors and traces completely; identify the first incorrect state, not only the final failure.
- Check recent changes, environment differences, inputs, and component boundaries.
- Compare with a nearby working path when one exists.
- State one falsifiable root-cause hypothesis and test it with the smallest useful experiment.
- Fix the source of the bad state, avoiding unrelated refactoring.
- Add regression coverage when it provides durable value, then run targeted and relevant broader verification.
Escalation
If repeated attempts fail, stop stacking changes. Reassess the evidence, assumptions, and architecture. After three materially different failed fixes, surface the architectural concern before another broad change.
Boundaries
- Do not require a formal investigation document for simple failures.
- Do not use root-cause language to delay an obvious, evidenced correction.
- Do not change multiple independent variables in one diagnostic experiment.
- Do not claim a fix from code inspection alone when the symptom can be tested.
- Do not require another skill; choose tests and review proportionate to the confirmed risk.
What ships with it: 1 file
230 B alongside SKILL.md
agents/
- openai.yaml230 B