Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Builds the gauntlet knowledge base from AST extraction and AI enrichment. Use when initializing or refreshing codebase knowledge for challenges.
.claude/skills/athola-extract/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | -20% | 0% |
| case-14 | ✗→✓ | ▲ Improved | -60% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -50% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -23% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -46% | 0% |
Build or rebuild the .gauntlet/knowledge.json knowledge base.
gauntlet:curate)gauntlet:graph-build)or a user-specified path
bash python3 ${CLAUDE_PLUGIN_ROOT}/scripts/extractor.py <target-dir>
detailfield with natural language explanation of business logic, data flow, architectural role, and rationale
matching imports, shared types, and data flow paths
in .gauntlet/annotations/
.gauntlet/knowledge.jsondistribution
.gauntlet/knowledge.json exists and is valid JSON after theskill completes; entries from .gauntlet/annotations/ are merged and not overwritten
(business_logic, architecture, data_flow, api_contract, pattern, dependency, error_handling) with coverage gaps identified
detail field containing a naturallanguage explanation (not just the raw AST node name)
modules sharing imports, shared types, or data flow paths
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-04 | fail→pass | 14,182 | 9,037 | -36% | 1 | 1 | 0% | 2,336 | 1,866 | -20% | 0 | 0 | — |
case-14 | fail→pass | 11,703 | 2,017 | -83% | 1 | 1 | 0% | 1,890 | 755 | -60% | 0 | 0 | — |
case-01 | fail→fail | 3,992 | 5,060 | +27% | 1 | 1 | 0% | 252 | 826 | +228% | 0 | 0 | — |
case-02 | fail→fail | 3,818 | 5,599 | +47% | 1 | 1 | 0% | 231 | 771 | +234% | 0 | 0 | — |
case-03 | fail→fail | 4,335 | 5,671 | +31% | 1 | 1 | 0% | 235 | 783 | +233% | 0 | 0 | — |
case-05 | fail→pass | 13,400 | 3,286 | -75% | 1 | 1 | 0% | 2,014 | 1,006 | -50% | 0 | 0 | — |
case-06 | pass→pass | 8,238 | 5,946 | -28% | 1 | 1 | 0% | 1,392 | 1,332 | -4% | 0 | 0 | — |
case-07 | fail→fail | 15,928 | 8,593 | -46% | 1 | 1 | 0% | 3,326 | 1,329 | -60% | 0 | 0 | — |
case-08 | fail→fail | 12,327 | 4,547 | -63% | 1 | 1 | 0% | 2,618 | 684 | -74% | 0 | 0 | — |
case-09 | fail→pass | 12,583 | 7,117 | -43% | 1 | 1 | 0% | 2,022 | 1,565 | -23% | 0 | 0 | — |
case-10 | fail→pass | 9,408 | 2,070 | -78% | 1 | 1 | 0% | 1,463 | 796 | -46% | 0 | 0 | — |
case-11 | fail→pass | 13,329 | 2,832 | -79% | 1 | 1 | 0% | 2,104 | 947 | -55% | 0 | 0 | — |
case-12 | fail→pass | 5,439 | 1,335 | -75% | 1 | 1 | 0% | 757 | 636 | -16% | 0 | 0 | — |
case-13 | fail→pass | 11,462 | 1,918 | -83% | 1 | 1 | 0% | 1,604 | 728 | -55% | 0 | 0 | — |
case-15 | fail→pass | 9,370 | 1,457 | -84% | 1 | 1 | 0% | 1,403 | 636 | -55% | 0 | 0 | — |
case-16 | pass→pass | 8,231 | 2,359 | -71% | 1 | 1 | 0% | 1,336 | 818 | -39% | 0 | 0 | — |
case-17 | fail→pass | 13,088 | 2,486 | -81% | 1 | 1 | 0% | 2,356 | 851 | -64% | 0 | 0 | — |
case-18 | fail→fail | 2,289 | 2,780 | +21% | 1 | 1 | 0% | 345 | 839 | +143% | 0 | 0 | — |
case-19 | pass→fail | 12,866 | 2,218 | -83% | 1 | 1 | 0% | 2,081 | 807 | -61% | 0 | 0 | — |
case-20 | fail→pass | 9,811 | 3,635 | -63% | 1 | 1 | 0% | 1,436 | 1,120 | -22% | 0 | 0 | — |
case-21 | fail→fail | 9,790 | 1,382 | -86% | 1 | 1 | 0% | 1,619 | 666 | -59% | 0 | 0 | — |
case-22 | pass→pass | 11,683 | 2,512 | -78% | 1 | 1 | 0% | 1,853 | 849 | -54% | 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 17 counted toward the lift figure. The other 5 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 +45 percentage points is the difference between those two pass rates over the 17 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.