Content
75%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
A lean, well-structured instruction-only skill: it names the exact tools, call syntaxes, and gating conditions for composing mini-apps, grounds them in a worked example, and wastes almost no tokens. The only meaningful gaps are the absence of error-recovery/validation checkpoints in the sibling-call workflow and a couple of near-duplicate gating conditions that could be tightened.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and imperative with essentially no padding of concepts Claude already knows — e.g., "Prefer many one-job apps in a workspace over one oversized app" and a compact artifact field list. Two spots could be trimmed: the "peer-exclusive data or capability" gating condition is repeated nearly verbatim across two bullets, and the stale-catalog rationale ("it reads as authoritative while pointing at capabilities that moved or vanished") is slightly long. That puts it at 'Efficient; minor instances of over-explanation that could be trimmed' — short of level 5's every-token-earns-its-place, well above level 3. | 4 / 5 |
Actionability | Guidance is concrete and executable in its instruction-only form: named tools with call syntax (`describe-workspace-apps` with `app: "<id>"`, `call-agent`, `invokeAgent()`, `provider-api-request`, `stageAs`, `query-staged-dataset`, `GET /_agent-native/agents?selfAppId=<app-id>`), a copy-paste-ready artifact example (`{ artifactType: "deal-set", artifactId: "hubspot-pipeline:deal-set:2026-06-18" }`), and a fully worked four-app example table with a flow. Per the rubric's code_vs_instruction note, absence of code is not penalized when guidance is this actionable. Level 5 is not earned because minor gaps remain — no example invocation payload for a sibling call and no guidance on creating/registering a new mini-app — keeping it at 'concrete code or commands with minor gaps'. | 4 / 5 |
Workflow Clarity | The discovery-to-invocation sequence is clearly staged — runtime `<available-apps>` block built from `discoverAgents()` → `describe-workspace-apps` for actual capability → `call-agent`/`invokeAgent()` — with explicit gating conditions at each step ("Use it only when the requested outcome depends on peer-exclusive data or capability..."), and the Example section walks a complete end-to-end flow (deal-brief asks hubspot-pipeline, gong-evidence, knowledge-base, then synthesizes). This is 'Clear sequence with most checkpoints present; minor validation gaps': there are no error-recovery checkpoints (e.g., what to do when a peer's agent card is unreachable or a sibling call fails), which keeps it below level 5; it stays above level 3 because the sequence and gating are explicit, and the operations are not destructive or batch operations that would trigger the level-3 cap. | 4 / 5 |
Progressive Disclosure | The body is well organized into coherent sections (Rule, Shape, Discovery And Invocation, Artifact Handoff, Provider APIs, Example, Don't) and closes with a one-level-deep, clearly signaled Related Skills list (a2a-protocol, actions, external-agents, storing-data). No bundle files exist (references/, scripts/, assets/ are absent) and none are needed, so navigation is easy and nothing is buried. It falls at 'Good structure; most content is appropriately placed' rather than level 5 because the skill is ~134 lines of inlined material — the artifact schema and provider-API detail would be natural candidates for a reference file if the skill grows — and the under-50-line simple-skill exception for a 5 does not apply; well above level 3, whose examples inline content that clearly belongs in separate files. | 4 / 5 |
Total | 16 / 20 Passed |