Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Bridge material for Layer-2 Hermes packaging.
.claude/skills/keith-leung-hermes-skill-dictionary-search-e3/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | -42% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 15% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 3% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -11% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -27% | 0% |
Bridge material for Layer-2 Hermes packaging. This skill shells out to an MCP client that talks to E's FastMCP server.
Use this skill when the agent needs to search a knowledge base via the capability-plane's dictionary-then-search MCP tool. The agent never calls HTTP directly; it invokes this skill, which calls the MCP client.
bashpython scripts/call_mcp_tool.py \ --server-cmd "python -m mcp_capability_plane.server.serve" \ --tool dictionary_search \ --params '{"query":"cozy cafe near downtown","domain":"dining","limit":5}'
query: natural language search intentdomain: abstract domain hint (dining / travel / mobility / general)limit: max results (1–50)JSON with data.items, data.total, data.domain, data.query.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 19,732 | 12,493 | -37% | 1 | 1 | 0% | 1,321 | 2,185 | +65% | 0 | 0 | — |
case-02 | fail→pass | 7,273 | 3,052 | -58% | 1 | 1 | 0% | 1,547 | 891 | -42% | 0 | 0 | — |
case-03 | fail→pass | 8,434 | 9,028 | +7% | 1 | 1 | 0% | 1,768 | 2,036 | +15% | 0 | 0 | — |
case-04 | fail→pass | 5,156 | 2,953 | -43% | 1 | 1 | 0% | 910 | 935 | +3% | 0 | 0 | — |
case-05 | fail→fail | 6,565 | 2,691 | -59% | 1 | 1 | 0% | 1,123 | 834 | -26% | 0 | 0 | — |
case-06 | fail→pass | 5,619 | 2,812 | -50% | 1 | 1 | 0% | 970 | 865 | -11% | 0 | 0 | — |
case-07 | fail→pass | 6,455 | 3,161 | -51% | 1 | 1 | 0% | 1,229 | 895 | -27% | 0 | 0 | — |
case-08 | fail→pass | 6,357 | 2,972 | -53% | 1 | 1 | 0% | 1,129 | 893 | -21% | 0 | 0 | — |
case-09 | pass→pass | 11,177 | 1,719 | -85% | 1 | 1 | 0% | 2,183 | 571 | -74% | 0 | 0 | — |
case-10 | pass→pass | 7,003 | 2,087 | -70% | 1 | 1 | 0% | 1,198 | 665 | -44% | 0 | 0 | — |
case-11 | pass→pass | 7,146 | 1,408 | -80% | 1 | 1 | 0% | 1,437 | 482 | -66% | 0 | 0 | — |
case-12 | fail→pass | 8,542 | 2,652 | -69% | 1 | 1 | 0% | 1,511 | 824 | -45% | 0 | 0 | — |
case-13 | fail→pass | 9,543 | 6,647 | -30% | 1 | 1 | 0% | 1,554 | 1,570 | +1% | 0 | 0 | — |
case-14 | fail→pass | 10,720 | 10,438 | -3% | 1 | 1 | 0% | 1,896 | 2,249 | +19% | 0 | 0 | — |
case-15 | fail→pass | 5,678 | 2,120 | -63% | 1 | 1 | 0% | 978 | 691 | -29% | 0 | 0 | — |
case-16 | fail→pass | 7,213 | 2,134 | -70% | 1 | 1 | 0% | 1,294 | 757 | -41% | 0 | 0 | — |
case-17 | fail→pass | 4,895 | 2,723 | -44% | 1 | 1 | 0% | 842 | 714 | -15% | 0 | 0 | — |
case-18 | pass→pass | 9,575 | 2,170 | -77% | 1 | 1 | 0% | 1,569 | 759 | -52% | 0 | 0 | — |
case-19 | pass→pass | 6,511 | 1,531 | -76% | 1 | 1 | 0% | 1,044 | 540 | -48% | 0 | 0 | — |
case-20 | pass→pass | 2,399 | 2,506 | +4% | 1 | 1 | 0% | 462 | 749 | +62% | 0 | 0 | — |
case-21 | pass→pass | 2,576 | 2,360 | -8% | 1 | 1 | 0% | 483 | 804 | +66% | 0 | 0 | — |
case-22 | pass→pass | 2,737 | 5,686 | +108% | 1 | 1 | 0% | 509 | 759 | +49% | 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. The headline lift of +55 percentage points is the difference between those two pass rates over the 22 comparable cases.
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.