planning-with-files
Optional persistent file-based planning for multi-phase or long-running work that benefits from recovery across sessions or…
General software bug diagnosis for broken, failing, throwing, flaky, or incorrect behavior. Read-only by default. Inside a requested fix, engineering-workflow owns the lifecycle. Use performance-forensics for latency, CPU, throughput, I/O, contention, or traces.
$ npx -y skills add khaneliman/khanelinix --skill diagnosing-bugs --agent claude-codeHow it fires
How this skill gets triggered: by you, by Claude, or both.
/diagnosing-bugsContext preview
The summary Claude sees to decide when to auto-load this skill.
General software bug diagnosis for broken, failing, throwing, flaky, or incorrect behavior. Read-only by default. Inside a requested fix, engineering-workflow owns the lifecycle. Use performance-forensics for latency, CPU, throughput, I/O, contention, or traces.
name: diagnosing-bugs description: "General software bug diagnosis for broken, failing, throwing, flaky, or incorrect behavior. Read-only by default. Inside a requested fix, engineering-workflow owns the lifecycle. Use performance-forensics for latency, CPU, throughput, I/O, contention, or traces." license: Complete terms in LICENSE
Use this skill to establish why general software behavior fails. Diagnosis-only work is read-only by default. If the user requests a fix, use this method inside the `engineering-workflow` Ground phase and return control to that lifecycle.
or change external state during diagnosis-only work without explicit authority.
endpoint or interaction contract is non-mutating.
temporary directories.
`[DEBUG-<id>]` prefix and remove it before handoff.
Choose one evidence path. Read [diagnosis-loop.md](references/diagnosis-loop.md) for the selected path's detail:
1. **Reproduction, preferred:** run a focused command that detects the user's exact symptom. Reuse it to test the cause and verify a requested fix. Minimize the reproduction only as far as needed to isolate the relevant uncertainty. 2. **Source or contract evidence:** when the defect follows conclusively from source, a schema, documentation, or an input/output contract, establish the mismatch directly. State what this proves and what still needs runtime confirmation. Do not claim an unexecuted reproduction or fix check.
Start with the best-supported explanation and a check that could disprove it. Compare alternative hypotheses when the evidence is ambiguous or a check fails; do not invent a quota of alternatives for an established mismatch.
A caller may strengthen either path, for example by requiring runtime reproduction for a timing bug or release acceptance. Honor that requirement without imposing it on every diagnosis. When neither path provides sufficient evidence, report the gap and request only the access or artifact needed to resolve it. Do not replace missing evidence with speculation.
For diagnosis-only work, stop after the evidence report. For a requested fix, return the evidence and supported cause to `engineering-workflow`. That lifecycle owns implementation, regression verification, review, and handoff.
Run packages directly with: Or install from the packages output. For example:
Optional persistent file-based planning for multi-phase or long-running work that benefits from recovery across sessions or…
Audit coding-agent sessions and mine cross-harness history for recurring corrections, skill-routing gaps, and harness faults. Use…
Design and audit agent instructions, skills, plugins, subagents, prompts, hooks, and provider configuration. Use for AI-tool…
Sketch interfaces, state, and module structure before code. Use for explicit design-led implementation or as a caller-owned…
Spawn N parallel candidates at the same task, pick a base, graft the strongest parts of the losers into it. Use for /arena,…
Bevy workflows for ECS, migrations, first-party UI/input/rendering/assets/shaders, plugin selection, scenes, scheduling, hot…