Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when starting work in a TermCanvas-managed repo to route between direct work, Hydra, or a narrow TermCanvas skill.
.claude/skills/blueberrycongee-using-termcanvas/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-08 | ✗→✓ | ▲ Improved | -63% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 31% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -6% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -1% | 0% |
| case-12 | ✗→✓ | ▲ Improved | -55% | 0% |
Route first. Choose the lightest path that preserves correctness.
challenge.investigate.security-audit.code-review.qa.agent, do it directly. Do not invoke Hydra by default.
or a staged workflow, use hydra.
instructions via hydra init-repo or the TermCanvas Hydra enable action.
Lead-driven, decision-point oriented. The Lead reads the codebase, picks the strategy, and dispatches workers for the steps that need a fresh agent process. Roles available: lead (the decider — not itself dispatched), dev (writes code AND its tests), reviewer (independent cross-model check). No separate researcher — the Lead does its own research. No separate tester — dev owns its own test surface.
hydra init --intent "..." --repo . then dispatch dev -> reviewerfor ambiguous, risky, or PRD-driven work
hydra watch after each dispatch to wait for the decision pointhydra spawn --task "..." --repo . for a single isolated workerfull Lead-driven loop
hydra spawn --task "..." --repo .spawned workers are actually involved.
termcanvas terminal create --prompt "..." rather than termcanvas terminal input.
hydra dispatch, immediately start hydra watch — do not ask whether to watch.hydra watch / hydra status / hydra ledger / hydra list --workflowsfor workflows created by hydra init.
hydra list and hydra cleanup <agentId> for direct workers created byhydra spawn.
When the session context contains a <memory-graph> block from TermCanvas:
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | pass→fail | 24,975 | 5,264 | -79% | 1 | 1 | 0% | 4,207 | 991 | -76% | 0 | 0 | — |
case-02 | fail→fail | 9,885 | 11,586 | +17% | 1 | 1 | 0% | 982 | 2,320 | +136% | 0 | 0 | — |
case-03 | fail→fail | 3,113 | 4,109 | +32% | 1 | 1 | 0% | 159 | 1,059 | +566% | 0 | 0 | — |
case-04 | pass→pass | 20,274 | 23,808 | +17% | 1 | 1 | 0% | 3,110 | 4,276 | +37% | 0 | 0 | — |
case-05 | fail→fail | 4,784 | 7,058 | +48% | 1 | 1 | 0% | 807 | 1,155 | +43% | 0 | 0 | — |
case-06 | fail→fail | 7,737 | 3,909 | -49% | 1 | 1 | 0% | 1,204 | 1,348 | +12% | 0 | 0 | — |
case-07 | fail→fail | 4,966 | 1,877 | -62% | 1 | 1 | 0% | 359 | 904 | +152% | 0 | 0 | — |
case-08 | fail→pass | 15,225 | 1,799 | -88% | 1 | 1 | 0% | 2,763 | 1,026 | -63% | 0 | 0 | — |
case-09 | fail→pass | 4,378 | 1,734 | -60% | 1 | 1 | 0% | 741 | 969 | +31% | 0 | 0 | — |
case-10 | fail→pass | 6,638 | 1,506 | -77% | 1 | 1 | 0% | 1,083 | 1,013 | -6% | 0 | 0 | — |
case-11 | fail→pass | 6,415 | 2,068 | -68% | 1 | 1 | 0% | 1,074 | 1,061 | -1% | 0 | 0 | — |
case-12 | fail→pass | 11,892 | 1,458 | -88% | 1 | 1 | 0% | 2,073 | 935 | -55% | 0 | 0 | — |
case-13 | fail→pass | 13,662 | 2,062 | -85% | 1 | 1 | 0% | 2,419 | 1,106 | -54% | 0 | 0 | — |
case-14 | pass→pass | 4,777 | 1,781 | -63% | 1 | 1 | 0% | 705 | 1,008 | +43% | 0 | 0 | — |
case-15 | fail→fail | 10,792 | 2,379 | -78% | 1 | 1 | 0% | 1,729 | 1,142 | -34% | 0 | 0 | — |
case-16 | pass→pass | 6,178 | 1,675 | -73% | 1 | 1 | 0% | 990 | 950 | -4% | 0 | 0 | — |
case-17 | pass→pass | 8,814 | 1,843 | -79% | 1 | 1 | 0% | 1,414 | 1,064 | -25% | 0 | 0 | — |
case-18 | pass→pass | 11,774 | 1,342 | -89% | 1 | 1 | 0% | 1,837 | 915 | -50% | 0 | 0 | — |
case-19 | pass→pass | 12,287 | 2,958 | -76% | 1 | 1 | 0% | 1,784 | 1,193 | -33% | 0 | 0 | — |
case-20 | pass→pass | 7,315 | 2,509 | -66% | 1 | 1 | 0% | 1,078 | 1,096 | +2% | 0 | 0 | — |
case-21 | pass→fail | 14,575 | 3,876 | -73% | 1 | 1 | 0% | 2,256 | 929 | -59% | 0 | 0 | — |
case-22 | pass→pass | 2,197 | 2,079 | -5% | 1 | 1 | 0% | 309 | 1,012 | +228% | 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 19 counted toward the lift figure. The other 3 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 19 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.