agentsclimarketplace

Bug investigation

Skill lklimek/claudius/skills/bug-investigation

Use when investigating a reported bug, diagnosing a failure, or doing root-cause analysis — especially before concluding "not a bug", tracing a symptom, or explaining unexpected behavior. Preloaded on investigation agents.From its SKILL.md

Install
npx -y skills add lklimek/claudius --skill bug-investigation

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 1 stars1 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.

SKILL.md

3.0 KB, 615 tokens by cl100k_base, as published. Nobody here has run it

Bug Investigation

Default discipline for diagnosing a reported bug. The user's reproducible observation is ground truth — the analysis must explain it, not explain it away.

Rules

  1. Observation over theory — the user's reproducible facts are ground truth to explain; their causal explanation is only a hypothesis. Refuting the theory ≠ explaining the observation. (User imprecision usually lives in the explanation, not the observation.)
  2. Entry point over name — statically trace the call graph from the actual UI/CLI entry point (the thing the user clicked/ran) down to the outcome. Never anchor on a function whose name merely matches the feature.
  3. The exercised path, not a correct path — when ≥2 plausible code paths exist, verify the one actually hit. Proving a correct path exists is not proving the user's path is correct.
  4. Reproduce or it's unsolved — if your analysis cannot reproduce the user's concrete observation, it is INCOMPLETE. Never conclude "not a bug" until the observation is explained. A clash between your analysis and a user-observed fact is a STOP signal, not a footnote. When the repro artifact is a test, it must be RED against the documented contract on the buggy revision, not green mirroring the defect — see coding-best-practices § Workflow Discipline TDD "Repro tests go RED first".
  5. Brief with the literal reproduction (coordinator-facing) — investigation spawn prompts MUST quote the user's exact reproduction steps and the literal entry point, and require: "trace from this entry point; if you can't reproduce the observed symptom, you haven't found the cause."

Failure Mode (worked example)

A real funds-safety bug (a receive address derived past the SPV gap window → invisible funds) was wrongly cleared as "not a bug"; only a user-supplied on-chain reproduction caught it.

Rule violatedWhat went wrong
Rule 1Anchored on the user's gap-limit THEORY instead of the OBSERVATION (button → index 32 → funds missing)
Rule 2Traced next_receive_address (a correct backend path) by name instead of from the UI button → add_receiving_address → legacy Wallet::receive_address
Rule 3Proved a correct path exists rather than checking the path the button actually calls
Rule 4Noticed an index 32 vs claimed 0 contradiction and rationalized it away instead of stopping

Composition

  • Complements coding-best-practices § Cross-Cutting Rules "Verify facts before acting on broad instructions" (that rule is about broad user directives; this skill is about not concluding prematurely during diagnosis).
  • This is a FORWARD trace (entry point → symptom); for the BACKWARD direction (which callers a changed function breaks) see grumpy-review's references/call-tree-walk.md.

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Keep looking

Skills are one crate of 325,949. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.