Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Debug systematically with root-cause analysis before fixes. Use for bugs, test failures, unexpected behavior, performance issues, CI failures, or system investigation.
.claude/skills/withkynam-vc-debug/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-13 | ✗→✓ | ▲ Improved | -7% | 0% |
| case-14 | ✗→✓ | ▲ Improved | 62% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 150% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 4% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 56% | 0% |
> Output style: Follow process/development-protocols/communication-standards.md — answer-first, plain language, no unexplained jargon, TL;DR on long responses.
Comprehensive framework combining systematic debugging, root cause tracing, defense-in-depth validation, verification protocols, and system-level investigation (logs, CI/CD, databases, performance).
NO FIXES WITHOUT ROOT CAUSE INVESTIGATION FIRST
Random fixes waste time and create new bugs. Find root cause, fix at source, validate at every layer, verify before claiming success.
Code-level: Test failures, bugs, unexpected behavior, build failures, integration problems System-level: Server errors, CI/CD pipeline failures, performance degradation, database issues, log analysis Always: Before claiming work complete
references/systematic-debugging.md)Four-phase framework: Root Cause Investigation → Pattern Analysis → Hypothesis Testing → Implementation. Complete each phase before proceeding. No fixes without Phase 1.
Load when: Any bug/issue requiring investigation and fix
references/root-cause-tracing.md)Trace bugs backward through call stack to find original trigger. Fix at source, not symptom. Includes scripts/find-polluter.sh for bisecting test pollution.
Load when: Error deep in call stack, unclear where invalid data originated
references/defense-in-depth.md)Validate at every layer: Entry validation → Business logic → Environment guards → Debug instrumentation
Load when: After finding root cause, need comprehensive validation
references/verification.md)Iron law: NO COMPLETION CLAIMS WITHOUT FRESH VERIFICATION EVIDENCE. Run command. Read output. Then claim result.
Load when: About to claim work complete, fixed, or passing
references/investigation-methodology.md)Five-step structured investigation for system-level issues: Initial Assessment → Data Collection → Analysis → Root Cause ID → Solution Development
Load when: Server incidents, system behavior analysis, multi-component failures
references/log-and-ci-analysis.md)Collect and analyze logs from servers, CI/CD pipelines (GitHub Actions), application layers. Tools: gh CLI, structured log queries, correlation across sources.
Load when: CI/CD pipeline failures, server errors, deployment issues
references/performance-diagnostics.md)Identify bottlenecks, analyze query performance, develop optimization strategies. Covers database queries, API response times, resource utilization.
Load when: Performance degradation, slow queries, high latency, resource exhaustion
references/reporting-standards.md)Structured diagnostic reports: Executive Summary → Technical Analysis → Recommendations → Evidence
Load when: Need to produce investigation report or diagnostic summary
references/task-management-debugging.md)Track investigation pipelines via Claude Native Tasks (TaskCreate, TaskUpdate, TaskList). Hydration pattern for multi-step investigations with dependency chains and parallel evidence collection. Fallback: Task tools are CLI-only — if unavailable (VSCode extension), use TodoWrite for tracking. Debug workflow remains fully functional.
Load when: Multi-component investigation (3+ steps), parallel log collection, coordinating debugger subagents
references/frontend-verification.md)Visual verification of frontend implementations via Chrome MCP (Claude Chrome Extension) or vc-agent-browser skill fallback. Detect if frontend-related → check Chrome MCP availability → screenshot + console error check → report. Skip if not frontend.
Load when: Implementation touches frontend files (tsx/jsx/vue/svelte/html/css), UI bugs, visual regressions
Code bug → systematic-debugging.md (Phase 1-4)
Deep in stack → root-cause-tracing.md (trace backward)
Found cause → defense-in-depth.md (add layers)
Claiming done → verification.md (verify first)
System issue → investigation-methodology.md (5 steps)
CI/CD failure → log-and-ci-analysis.md
Slow system → performance-diagnostics.md
Need report → reporting-standards.md
Frontend fix → frontend-verification.md (Chrome/devtools)sqlite3 CLI and drizzle-kit studio for SQLite/libSQL diagnosticsgh CLI for GitHub Actions logs and pipeline debuggingvc-docs-seeker skill for package/plugin docs; vc-scout skill for codebase exploration/vc-scout or /vc-scout ext for finding relevant filesvc-agent-browser skill for visual verification (screenshots, console, network)vc-problem-solving skill when stuck on complex issuesStop and follow process if thinking:
All mean: Return to systematic process.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-13 | fail→pass | 15,523 | 7,466 | -52% | 1 | 1 | 0% | 2,600 | 2,415 | -7% | 0 | 0 | — |
case-14 | fail→pass | 7,574 | 5,523 | -27% | 1 | 1 | 0% | 1,279 | 2,066 | +62% | 0 | 0 | — |
case-02 | pass→pass | 26,425 | 26,478 | +0% | 1 | 1 | 0% | 5,106 | 7,128 | +40% | 0 | 0 | — |
case-01 | fail→fail | 15,927 | 15,568 | -2% | 1 | 1 | 0% | 2,830 | 3,779 | +34% | 0 | 0 | — |
case-03 | pass→pass | 6,103 | 6,680 | +9% | 1 | 1 | 0% | 1,077 | 2,469 | +129% | 0 | 0 | — |
case-04 | pass→fail | 8,819 | 2,036 | -77% | 1 | 1 | 0% | 1,475 | 1,503 | +2% | 0 | 0 | — |
case-05 | fail→pass | 5,071 | 7,208 | +42% | 1 | 1 | 0% | 931 | 2,330 | +150% | 0 | 0 | — |
case-06 | pass→pass | 8,583 | 6,240 | -27% | 1 | 1 | 0% | 1,620 | 2,190 | +35% | 0 | 0 | — |
case-12 | fail→pass | 17,161 | 9,875 | -42% | 1 | 1 | 0% | 2,772 | 2,879 | +4% | 0 | 0 | — |
case-07 | fail→pass | 10,018 | 9,813 | -2% | 1 | 1 | 0% | 1,807 | 2,825 | +56% | 0 | 0 | — |
case-08 | fail→pass | 8,299 | 3,726 | -55% | 1 | 1 | 0% | 1,384 | 1,862 | +35% | 0 | 0 | — |
case-09 | pass→pass | 12,490 | 7,413 | -41% | 1 | 1 | 0% | 2,029 | 2,333 | +15% | 0 | 0 | — |
case-10 | fail→fail | 8,032 | 6,536 | -19% | 1 | 1 | 0% | 1,220 | 2,294 | +88% | 0 | 0 | — |
case-11 | pass→pass | 11,197 | 3,899 | -65% | 1 | 1 | 0% | 2,080 | 1,931 | -7% | 0 | 0 | — |
case-15 | pass→pass | 7,371 | 3,636 | -51% | 1 | 1 | 0% | 1,204 | 1,692 | +41% | 0 | 0 | — |
case-16 | pass→pass | 12,846 | 5,893 | -54% | 1 | 1 | 0% | 1,951 | 2,227 | +14% | 0 | 0 | — |
case-17 | pass→pass | 12,438 | 7,804 | -37% | 1 | 1 | 0% | 2,195 | 2,632 | +20% | 0 | 0 | — |
case-18 | pass→pass | 14,352 | 4,586 | -68% | 1 | 1 | 0% | 2,289 | 2,078 | -9% | 0 | 0 | — |
case-19 | fail→pass | 7,252 | 4,233 | -42% | 1 | 1 | 0% | 1,133 | 1,806 | +59% | 0 | 0 | — |
case-20 | pass→pass | 18,368 | 12,773 | -30% | 1 | 1 | 0% | 2,929 | 3,335 | +14% | 0 | 0 | — |
case-21 | fail→pass | 14,696 | 7,784 | -47% | 1 | 1 | 0% | 2,389 | 2,432 | +2% | 0 | 0 | — |
case-22 | pass→pass | 10,360 | 8,029 | -23% | 1 | 1 | 0% | 1,675 | 2,346 | +40% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +32 percentage points is the difference between those two pass rates over the 22 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.