Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Systematically debug issues using graph-powered code navigation
.claude/skills/tirth8205-debug-issue/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-05 | ✗→✓ | ▲ Improved | 22% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -23% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -34% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -45% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -26% | 0% |
Trace a bug through the knowledge graph before reading source.
semantic_search_nodes_tool to find code related to the issue.query_graph_tool with callers_of and callees_of to trace the call chain in both directions.get_flow_tool for the execution path that reaches the suspect code. Its entry point is where the bug is triggered.detect_changes_tool to check whether a recent change caused the issue.get_impact_radius_tool on the suspect files to see what a fix would affect.get_minimal_context_tool(task="<your task>") before any other graph tool.detail_level="minimal" wherever a tool accepts it. Use "standard" only when minimal is not enough.query_graph_tool call over a broad listing call.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 31,912 | 17,760 | -44% | 1 | 1 | 0% | 5,306 | 646 | -88% | 0 | 0 | — |
case-02 | fail→fail | 31,102 | 24,198 | -22% | 1 | 1 | 0% | 4,215 | 594 | -86% | 0 | 0 | — |
case-03 | fail→fail | 21,339 | 17,409 | -18% | 1 | 1 | 0% | 3,491 | 614 | -82% | 0 | 0 | — |
case-04 | fail→fail | 29,155 | 12,720 | -56% | 1 | 1 | 0% | 1,923 | 742 | -61% | 0 | 0 | — |
case-05 | fail→pass | 15,830 | 24,203 | +53% | 1 | 1 | 0% | 1,467 | 1,792 | +22% | 0 | 0 | — |
case-06 | fail→fail | 18,761 | 17,910 | -5% | 1 | 1 | 0% | 1,794 | 615 | -66% | 0 | 0 | — |
case-07 | fail→pass | 22,222 | 15,633 | -30% | 1 | 1 | 0% | 2,544 | 1,952 | -23% | 0 | 0 | — |
case-08 | fail→pass | 16,187 | 14,716 | -9% | 1 | 1 | 0% | 1,496 | 984 | -34% | 0 | 0 | — |
case-09 | fail→pass | 15,033 | 6,676 | -56% | 1 | 1 | 0% | 2,161 | 1,185 | -45% | 0 | 0 | — |
case-10 | fail→pass | 17,325 | 12,597 | -27% | 1 | 1 | 0% | 1,873 | 1,394 | -26% | 0 | 0 | — |
case-11 | fail→pass | 20,263 | 8,382 | -59% | 1 | 1 | 0% | 1,965 | 1,560 | -21% | 0 | 0 | — |
case-12 | fail→pass | 14,778 | 16,558 | +12% | 1 | 1 | 0% | 1,396 | 1,146 | -18% | 0 | 0 | — |
case-13 | fail→pass | 24,830 | 5,741 | -77% | 1 | 1 | 0% | 2,730 | 1,200 | -56% | 0 | 0 | — |
case-14 | fail→fail | 10,674 | 8,938 | -16% | 1 | 1 | 0% | 1,407 | 868 | -38% | 0 | 0 | — |
case-15 | pass→fail | 23,585 | 15,263 | -35% | 1 | 1 | 0% | 1,372 | 551 | -60% | 0 | 0 | — |
case-16 | pass→pass | 11,634 | 8,010 | -31% | 1 | 1 | 0% | 797 | 605 | -24% | 0 | 0 | — |
case-17 | fail→fail | 18,171 | 18,686 | +3% | 1 | 1 | 0% | 2,243 | 630 | -72% | 0 | 0 | — |
case-18 | fail→pass | 16,262 | 14,767 | -9% | 1 | 1 | 0% | 994 | 1,321 | +33% | 0 | 0 | — |
case-19 | pass→fail | 16,970 | 14,974 | -12% | 1 | 1 | 0% | 2,102 | 450 | -79% | 0 | 0 | — |
case-20 | pass→fail | 18,231 | 16,672 | -9% | 1 | 1 | 0% | 2,124 | 477 | -78% | 0 | 0 | — |
case-21 | pass→fail | 24,485 | 16,311 | -33% | 1 | 1 | 0% | 2,974 | 525 | -82% | 0 | 0 | — |
case-22 | fail→pass | 16,255 | 12,990 | -20% | 1 | 1 | 0% | 1,660 | 1,410 | -15% | 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, and 12 counted toward the lift figure. The other 10 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +27 percentage points is the difference between those two pass rates over the 12 comparable cases. 4 cases got worse with the skill loaded, and they are 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.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 8/24/2026 | +23% |
| gemini-3.6-flash | verified | 8/9/2026 | +36% |
Other measured skills in the registry, with their headline benchmark lift.