Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Searches the code knowledge graph by function, class, or type using FTS5 full-text search. Use when locating code entities by name or qualified path.
.claude/skills/athola-graph-search/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | -58% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -43% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -40% | 0% |
| case-20 | ✗→✓ | ▲ Improved | 33% | 0% |
| case-14 | ✓→✗ | ▼ Worse | -74% | 0% |
Search .gauntlet/graph.db for code entities by name.
gauntlet:graph-build)pensive:blast-radius)bash python3 ${CLAUDE_PLUGIN_ROOT}/scripts/graph_query.py \ --action search --query "<term>" --limit 20
Optional filters:
--kind Function to search only functions--kind Class to search only classesline numbers, and relevance score for each match.
top result's source file.
The search engine detects query patterns:
UserService): boosts Classand Type results
get_users): boosts Functionresults
app.models.User): boostsqualified name matches
The graph must be built first. If .gauntlet/graph.db does not exist, suggest running the graph-build skill.
.gauntlet/graph.db does not exist, the skill surfacesthe missing-prerequisite error and suggests gauntlet:graph-build rather than failing silently
relevance score for each match (up to --limit results)
Class/Type results, snake_case boosts Function results, dotted path boosts qualified name matches
results are shown
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-07 | fail→fail | 5,780 | 6,212 | +7% | 1 | 1 | 0% | 345 | 779 | +126% | 0 | 0 | — |
case-19 | fail→fail | 16,418 | 5,579 | -66% | 1 | 1 | 0% | 2,714 | 726 | -73% | 0 | 0 | — |
case-01 | fail→fail | 3,105 | 5,849 | +88% | 1 | 1 | 0% | 399 | 720 | +80% | 0 | 0 | — |
case-02 | fail→fail | 9,091 | 6,747 | -26% | 1 | 1 | 0% | 846 | 800 | -5% | 0 | 0 | — |
case-03 | fail→fail | 4,119 | 7,193 | +75% | 1 | 1 | 0% | 559 | 698 | +25% | 0 | 0 | — |
case-04 | fail→pass | 14,544 | 3,204 | -78% | 1 | 1 | 0% | 2,289 | 962 | -58% | 0 | 0 | — |
case-05 | fail→pass | 10,628 | 3,745 | -65% | 1 | 1 | 0% | 1,797 | 1,025 | -43% | 0 | 0 | — |
case-06 | fail→pass | 9,769 | 3,239 | -67% | 1 | 1 | 0% | 1,628 | 977 | -40% | 0 | 0 | — |
case-08 | fail→fail | 7,698 | 5,620 | -27% | 1 | 1 | 0% | 703 | 782 | +11% | 0 | 0 | — |
case-09 | fail→fail | 10,068 | 6,393 | -37% | 1 | 1 | 0% | 475 | 780 | +64% | 0 | 0 | — |
case-10 | fail→fail | 10,291 | 6,301 | -39% | 1 | 1 | 0% | 1,590 | 971 | -39% | 0 | 0 | — |
case-11 | fail→fail | 16,864 | 5,175 | -69% | 1 | 1 | 0% | 2,946 | 714 | -76% | 0 | 0 | — |
case-12 | fail→fail | 4,208 | 6,702 | +59% | 1 | 1 | 0% | 581 | 969 | +67% | 0 | 0 | — |
case-13 | fail→fail | 6,538 | 2,732 | -58% | 1 | 1 | 0% | 1,016 | 658 | -35% | 0 | 0 | — |
case-14 | pass→fail | 14,918 | 5,897 | -60% | 1 | 1 | 0% | 2,872 | 760 | -74% | 0 | 0 | — |
case-15 | pass→fail | 15,779 | 5,892 | -63% | 1 | 1 | 0% | 1,726 | 737 | -57% | 0 | 0 | — |
case-16 | fail→fail | 4,911 | 5,632 | +15% | 1 | 1 | 0% | 189 | 722 | +282% | 0 | 0 | — |
case-17 | fail→fail | 7,937 | 5,117 | -36% | 1 | 1 | 0% | 1,137 | 692 | -39% | 0 | 0 | — |
case-18 | fail→fail | 8,858 | 1,848 | -79% | 1 | 1 | 0% | 1,452 | 695 | -52% | 0 | 0 | — |
case-20 | fail→pass | 12,289 | 4,723 | -62% | 1 | 1 | 0% | 982 | 1,309 | +33% | 0 | 0 | — |
case-21 | fail→fail | 12,256 | 5,288 | -57% | 1 | 1 | 0% | 2,041 | 828 | -59% | 0 | 0 | — |
case-22 | pass→fail | 8,194 | 5,773 | -30% | 1 | 1 | 0% | 1,376 | 779 | -43% | 0 | 0 | — |
case-23 | fail→fail | 8,127 | 4,905 | -40% | 1 | 1 | 0% | 682 | 704 | +3% | 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. 23 cases were attempted, and 5 counted toward the lift figure. The other 18 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 +4 percentage points is the difference between those two pass rates over the 5 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.
Other measured skills in the registry, with their headline benchmark lift.