Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Load the AgentOps language and
.claude/skills/boshu2-domain/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -2% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -18% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -56% | 0% |
| case-12 | ✗→✓ | ▲ Improved | -13% | 0% |
| case-14 | ✗→✓ | ▲ Improved | -69% | 0% |
Use this read-only library when an AgentOps term or bounded-context boundary needs precise meaning.
Returning the exact contract definition works because two contexts that read the same source line cannot drift apart on a term; a paraphrase reintroduces the ambiguity the lookup exists to remove.
Named failure mode — synonym smuggling: answering with a near-synonym ("close" for "verdict") that quietly imports lifecycle authority the term does not carry.
Anti-pattern: defining a term from memory because the contract file is one read away. Corrective: always cite the definition and its source path from the live contract, even for terms you are sure about.
docs/contracts/ubiquitous-language.md for the term.docs/contracts/bounded-contexts.yaml only when ownership or a portboundary matters.
Do not invent synonyms that imply lifecycle authority. In particular, Plan, Candidate, manifest, verdict, revision, strategy, and adapter are semantic terms; queue, claim, lease, close, land, release, and delivery belong to caller systems rather than AgentOps core state.
The product-architecture terms — operations layer, federated integration graph, semantic work-and-proof protocol, RPI traversal — and their forbidden conflations live in the same contract. The traversal is RPI; the graph is the topology; the protocol defines interoperability; the operations layer is the product. None of the four is "a loop".
Vocabulary changes are normal source edits to the two contracts above. This skill does not promote terms, mutate a knowledge index, or create continuation.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 5,515 | 2,614 | -53% | 1 | 1 | 0% | 734 | 716 | -2% | 0 | 0 | — |
case-02 | fail→pass | 5,284 | 1,588 | -70% | 1 | 1 | 0% | 774 | 636 | -18% | 0 | 0 | — |
case-03 | fail→fail | 3,859 | 4,736 | +23% | 1 | 1 | 0% | 590 | 599 | +2% | 0 | 0 | — |
case-04 | fail→fail | 12,066 | 4,905 | -59% | 1 | 1 | 0% | 1,862 | 576 | -69% | 0 | 0 | — |
case-05 | fail→pass | 9,494 | 1,948 | -79% | 1 | 1 | 0% | 1,531 | 679 | -56% | 0 | 0 | — |
case-06 | fail→fail | 8,131 | 4,279 | -47% | 1 | 1 | 0% | 1,319 | 558 | -58% | 0 | 0 | — |
case-07 | fail→fail | 12,129 | 4,244 | -65% | 1 | 1 | 0% | 2,091 | 573 | -73% | 0 | 0 | — |
case-08 | fail→fail | 11,944 | 2,505 | -79% | 1 | 1 | 0% | 1,740 | 594 | -66% | 0 | 0 | — |
case-09 | fail→fail | 11,838 | 5,826 | -51% | 1 | 1 | 0% | 1,962 | 651 | -67% | 0 | 0 | — |
case-10 | fail→fail | 11,596 | 5,961 | -49% | 1 | 1 | 0% | 949 | 560 | -41% | 0 | 0 | — |
case-11 | fail→fail | 11,208 | 5,155 | -54% | 1 | 1 | 0% | 1,781 | 563 | -68% | 0 | 0 | — |
case-12 | fail→pass | 11,771 | 9,441 | -20% | 1 | 1 | 0% | 1,824 | 1,592 | -13% | 0 | 0 | — |
case-13 | fail→fail | 11,882 | 4,858 | -59% | 1 | 1 | 0% | 1,846 | 579 | -69% | 0 | 0 | — |
case-14 | fail→pass | 14,074 | 4,655 | -67% | 1 | 1 | 0% | 2,304 | 708 | -69% | 0 | 0 | — |
case-15 | fail→pass | 9,105 | 2,257 | -75% | 1 | 1 | 0% | 1,417 | 730 | -48% | 0 | 0 | — |
case-16 | fail→fail | 12,951 | 4,749 | -63% | 1 | 1 | 0% | 1,921 | 560 | -71% | 0 | 0 | — |
case-17 | fail→pass | 9,266 | 3,321 | -64% | 1 | 1 | 0% | 1,339 | 910 | -32% | 0 | 0 | — |
case-18 | pass→pass | 5,735 | 4,719 | -18% | 1 | 1 | 0% | 1,001 | 1,282 | +28% | 0 | 0 | — |
case-19 | pass→pass | 11,949 | 6,999 | -41% | 1 | 1 | 0% | 1,943 | 1,570 | -19% | 0 | 0 | — |
case-20 | pass→pass | 6,301 | 3,218 | -49% | 1 | 1 | 0% | 1,030 | 909 | -12% | 0 | 0 | — |
case-21 | fail→pass | 13,205 | 2,321 | -82% | 1 | 1 | 0% | 1,975 | 725 | -63% | 0 | 0 | — |
case-22 | fail→pass | 13,085 | 2,851 | -78% | 1 | 1 | 0% | 1,835 | 871 | -53% | 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 13 counted toward the lift figure. The other 9 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 +41 percentage points is the difference between those two pass rates over the 13 comparable cases. 4 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.