Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use Ontoly's deterministic Software Graph and MCP server for codebase architecture, request tracing, dependency analysis, and impact analysis.
.claude/skills/davepoon-ontoly-software-graph/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 64% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -45% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -27% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -48% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -27% | 0% |
Use Ontoly when a coding agent needs graph-backed software understanding before searching source files directly.
Ontoly builds a deterministic Software Graph from a TypeScript repository and exposes it through CLI queries, MCP capabilities, and agent skills. The skill teaches workflow only; Ontoly remains the source of truth.
.ontoly/, SoftwareGraph.json, or documented Ontoly scripts.ontoly build ..textUse Ontoly to explain this repository's architecture.
textUse Ontoly to trace the login flow from route to controller, service, and repository.
textUse Ontoly to find what depends on UserRepository and what breaks if it changes.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 4,115 | 5,734 | +39% | 1 | 1 | 0% | 274 | 768 | +180% | 0 | 0 | — |
case-02 | fail→pass | 6,960 | 9,375 | +35% | 1 | 1 | 0% | 1,202 | 1,968 | +64% | 0 | 0 | — |
case-03 | fail→fail | 5,686 | 2,706 | -52% | 1 | 1 | 0% | 310 | 741 | +139% | 0 | 0 | — |
case-04 | fail→pass | 15,983 | 6,333 | -60% | 1 | 1 | 0% | 2,655 | 1,455 | -45% | 0 | 0 | — |
case-05 | fail→pass | 11,850 | 5,865 | -51% | 1 | 1 | 0% | 1,950 | 1,428 | -27% | 0 | 0 | — |
case-06 | pass→pass | 13,095 | 9,412 | -28% | 1 | 1 | 0% | 2,066 | 1,955 | -5% | 0 | 0 | — |
case-07 | pass→pass | 12,717 | 6,789 | -47% | 1 | 1 | 0% | 2,061 | 1,593 | -23% | 0 | 0 | — |
case-08 | pass→fail | 15,649 | 3,044 | -81% | 1 | 1 | 0% | 3,055 | 846 | -72% | 0 | 0 | — |
case-13 | fail→fail | 3,600 | 2,552 | -29% | 1 | 1 | 0% | 627 | 806 | +29% | 0 | 0 | — |
case-09 | fail→pass | 14,158 | 5,187 | -63% | 1 | 1 | 0% | 2,563 | 1,338 | -48% | 0 | 0 | — |
case-10 | fail→pass | 10,614 | 5,667 | -47% | 1 | 1 | 0% | 1,975 | 1,443 | -27% | 0 | 0 | — |
case-11 | fail→fail | 8,568 | 2,657 | -69% | 1 | 1 | 0% | 1,455 | 845 | -42% | 0 | 0 | — |
case-12 | pass→pass | 11,661 | 5,366 | -54% | 1 | 1 | 0% | 2,035 | 1,295 | -36% | 0 | 0 | — |
case-14 | fail→fail | 9,779 | 7,000 | -28% | 1 | 1 | 0% | 1,689 | 1,535 | -9% | 0 | 0 | — |
case-15 | pass→pass | 12,680 | 8,715 | -31% | 1 | 1 | 0% | 1,974 | 1,882 | -5% | 0 | 0 | — |
case-16 | pass→pass | 14,012 | 10,942 | -22% | 1 | 1 | 0% | 2,689 | 1,955 | -27% | 0 | 0 | — |
case-17 | pass→pass | 10,267 | 5,464 | -47% | 1 | 1 | 0% | 1,584 | 1,392 | -12% | 0 | 0 | — |
case-18 | fail→pass | 12,330 | 4,405 | -64% | 1 | 1 | 0% | 2,134 | 1,149 | -46% | 0 | 0 | — |
case-19 | pass→pass | 8,634 | 3,550 | -59% | 1 | 1 | 0% | 1,518 | 1,027 | -32% | 0 | 0 | — |
case-20 | pass→pass | 2,957 | 1,847 | -38% | 1 | 1 | 0% | 503 | 737 | +47% | 0 | 0 | — |
case-21 | pass→pass | 5,201 | 2,868 | -45% | 1 | 1 | 0% | 919 | 951 | +3% | 0 | 0 | — |
case-22 | pass→fail | 6,229 | 6,743 | +8% | 1 | 1 | 0% | 1,086 | 1,600 | +47% | 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 20 counted toward the lift figure. The other 2 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 +18 percentage points is the difference between those two pass rates over the 20 comparable cases. 2 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.