Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →SOP wrapping vault_query_graph — traverse knowledge graph from a node to explore its neighborhood and context.
.claude/skills/yogsoth-ai-wiki-graph-query/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-05 | ✗→✓ | ▲ Improved | 53% | 0% |
| case-19 | ✗→✓ | ▲ Improved | -42% | 0% |
| case-21 | ✗→✓ | ▲ Improved | -41% | 0% |
| case-09 | ✓→✗ | ▼ Worse | -88% | 0% |
| case-17 | ✓→✗ | ▼ Worse | -94% | 0% |
Explore the neighborhood of a node in the knowledge graph. Used for understanding context before updates, finding connection opportunities, and identifying orphans.
<HARD-GATE> Node path must reference an existing file in the vault. Verify the file exists before calling vault_query_graph. </HARD-GATE>
vault_query_graph
| Param | Required | Description | |-------|----------|-------------| | node | yes | Starting node path (e.g., "concepts/attention.md") | | direction | yes | Traversal direction: "in", "out", or "both" | | edge_type | no | Filter by edge type | | depth | no | Traversal depth (default 1, max 3) |
vault_query_graph with node, direction, and optional filtersReturns: { nodes_found: number, edges_found: number, subgraph: object }
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-07 | fail→fail | 9,663 | 18,219 | +89% | 1 | 1 | 0% | 694 | 866 | +25% | 0 | 0 | — |
case-08 | fail→fail | 17,315 | 14,773 | -15% | 1 | 1 | 0% | 526 | 525 | -0% | 0 | 0 | — |
case-01 | fail→fail | 19,114 | 15,897 | -17% | 1 | 1 | 0% | 601 | 664 | +10% | 0 | 0 | — |
case-02 | fail→fail | 12,061 | 18,676 | +55% | 1 | 1 | 0% | 1,077 | 1,049 | -3% | 0 | 0 | — |
case-03 | fail→fail | 15,814 | 16,005 | +1% | 1 | 1 | 0% | 287 | 686 | +139% | 0 | 0 | — |
case-04 | fail→fail | 19,236 | 20,652 | +7% | 1 | 1 | 0% | 739 | 1,126 | +52% | 0 | 0 | — |
case-05 | fail→pass | 18,249 | 9,160 | -50% | 1 | 1 | 0% | 599 | 916 | +53% | 0 | 0 | — |
case-06 | fail→fail | 9,710 | 17,357 | +79% | 1 | 1 | 0% | 699 | 623 | -11% | 0 | 0 | — |
case-09 | pass→fail | 30,455 | 44,574 | +46% | 1 | 1 | 0% | 4,181 | 522 | -88% | 0 | 0 | — |
case-10 | fail→fail | 17,652 | 15,287 | -13% | 1 | 1 | 0% | 349 | 540 | +55% | 0 | 0 | — |
case-11 | fail→fail | 15,983 | 18,417 | +15% | 1 | 1 | 0% | 390 | 587 | +51% | 0 | 0 | — |
case-12 | fail→fail | 17,395 | 17,599 | +1% | 1 | 1 | 0% | 2,220 | 586 | -74% | 0 | 0 | — |
case-13 | fail→fail | 7,215 | 22,660 | +214% | 1 | 1 | 0% | 321 | 816 | +154% | 0 | 0 | — |
case-14 | fail→fail | 34,903 | 29,424 | -16% | 1 | 1 | 0% | 3,258 | 2,971 | -9% | 0 | 0 | — |
case-15 | fail→fail | 19,916 | 32,724 | +64% | 1 | 1 | 0% | 908 | 1,673 | +84% | 0 | 0 | — |
case-16 | pass→pass | 16,464 | 45,834 | +178% | 1 | 1 | 0% | 2,066 | 4,650 | +125% | 0 | 0 | — |
case-17 | pass→fail | 42,295 | 15,381 | -64% | 1 | 1 | 0% | 8,166 | 486 | -94% | 0 | 0 | — |
case-18 | fail→fail | 28,474 | 17,767 | -38% | 1 | 1 | 0% | 3,942 | 811 | -79% | 0 | 0 | — |
case-19 | fail→pass | 17,963 | 11,643 | -35% | 1 | 1 | 0% | 2,264 | 1,320 | -42% | 0 | 0 | — |
case-20 | fail→fail | 35,905 | 17,229 | -52% | 1 | 1 | 0% | 5,907 | 741 | -87% | 0 | 0 | — |
case-21 | fail→pass | 23,128 | 14,525 | -37% | 1 | 1 | 0% | 3,138 | 1,845 | -41% | 0 | 0 | — |
case-22 | fail→fail | 16,601 | 17,234 | +4% | 1 | 1 | 0% | 423 | 939 | +122% | 0 | 0 | — |
case-23 | fail→fail | 28,699 | 16,868 | -41% | 1 | 1 | 0% | 4,215 | 502 | -88% | 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 4 counted toward the lift figure. The other 19 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. A headline lift is not published for this run.
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.