---
name: systematic-debugging
description: ALWAYS use this when the user mentions any bug, flaky behavior, failing test, broken build, regression, runtime error, performance mystery, or unexpected output, especially when guessing at fixes feels tempting.
Systematic Debugging
Selective Reading Rule
Start with:
references/usage-routing.md
references/quality-checklist.md
Then load only the inherited docs, scripts, assets, or examples that match the user's actual task.
Selective Reading Rule
Start with:
references/senior-master-standard.md
references/evidence-capture.md
references/root-cause-tracing.md
Read both before proposing fixes when the problem crosses multiple layers, has already resisted one fix, or could be a symptom instead of the source.
Purpose
Turn debugging into evidence collection instead of guesswork.
The goal is not to try clever patches quickly. The goal is to isolate root cause, change the smallest thing that matters, and prove the result.
Use This Skill When
- a test, build, script, or app behavior is failing
- a bug appears intermittent or hard to explain
- the first fix did not work
- a multi-step system could be breaking at more than one boundary
Core Law
No fix without root-cause investigation first.
Workflow
- Reproduce.
- capture the exact failure, frequency, and boundary conditions
- Collect evidence.
- read errors fully, inspect logs, and add instrumentation where signal is missing
- Trace backwards.
- walk from the symptom to the first bad value, state transition, config, or side effect
- Change the narrowest thing.
- patch the cause, not the surface symptom
- Verify on purpose.
- rerun the failing path, then the nearest regression checks
Non-Negotiables
- Never describe a theory as a fact until the evidence supports it.
- Never hide uncertainty behind big confident language.
- Never stack speculative fixes together when one proving change would teach more.
- If reproduction is missing, say so explicitly and switch to evidence-gathering mode.
Deliverable
Close with:
- the reproduction or observation path
- the proven root cause
- the fix applied
- the commands or checks used to verify the result
1---2name: systematic-debugging3description: ---4---5---6name: systematic-debugging7description: ALWAYS use this when the user mentions any bug, flaky behavior, failing test, broken build, regression, runtime error, performance mystery, or unexpected output, especially when guessing at fixes feels tempting.8---910# Systematic Debugging1112## Selective Reading Rule1314Start with:1516- `references/usage-routing.md`17- `references/quality-checklist.md`1819Then load only the inherited docs, scripts, assets, or examples that match the user's actual task.2021## Selective Reading Rule2223Start with:2425- `references/senior-master-standard.md`26- `references/evidence-capture.md`27- `references/root-cause-tracing.md`2829Read both before proposing fixes when the problem crosses multiple layers, has already resisted one fix, or could be a symptom instead of the source.3031## Purpose3233Turn debugging into evidence collection instead of guesswork.3435The goal is not to try clever patches quickly. The goal is to isolate root cause, change the smallest thing that matters, and prove the result.3637## Use This Skill When3839- a test, build, script, or app behavior is failing40- a bug appears intermittent or hard to explain41- the first fix did not work42- a multi-step system could be breaking at more than one boundary4344## Core Law4546No fix without root-cause investigation first.4748## Workflow49501. Reproduce.51 - capture the exact failure, frequency, and boundary conditions522. Collect evidence.53 - read errors fully, inspect logs, and add instrumentation where signal is missing543. Trace backwards.55 - walk from the symptom to the first bad value, state transition, config, or side effect564. Change the narrowest thing.57 - patch the cause, not the surface symptom585. Verify on purpose.59 - rerun the failing path, then the nearest regression checks6061## Non-Negotiables6263- Never describe a theory as a fact until the evidence supports it.64- Never hide uncertainty behind big confident language.65- Never stack speculative fixes together when one proving change would teach more.66- If reproduction is missing, say so explicitly and switch to evidence-gathering mode.6768## Deliverable6970Close with:7172- the reproduction or observation path73- the proven root cause74- the fix applied75- the commands or checks used to verify the result