Content
96%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.
An exemplary operational skill body: lean and table-driven with no concept padding, copy-paste-ready commands and code, a fully sequenced workflow with dry-run gates, atomic writes, explicit verification, and tuning feedback loops. The single-file structure is well organized but slightly dense; moving the per-error-code CLI handling into a reference file would make the overview tighter.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and imperative throughout: a five-question discovery table, exact CLI invocations, a YAML example, and a wiring snippet, with zero explanation of concepts Claude already knows (no 'what is an embedding' or 'what is Redis' padding). The opening paragraph ('The Router embeds the message, finds the nearest example phrasings, groups them by intent, applies a threshold...') describes product-specific behavior Claude would not know, so it earns its tokens. Not 4: there is no identifiable passage of over-explanation that could be trimmed without losing operational information. | 5 / 5 |
Actionability | Fully executable guidance: copy-paste-ready CLI commands ('iris --json --no-input intent-router service create --name support --provider openai --model text-embedding-3-small --dimensions 1536 --role read --wait --write-env .env'), a concrete intents.yaml example, working Python ('router = IntentRouter.from_env()', 'found = router.match(message)'), and a framework wiring table naming exact replacement points (LangGraph router node, OpenAI Agents SDK triage agent, Google ADK before_agent_callback). The 'else: ... # question 4' placeholder is justified flexibility since the fallback is project-specific by design. Not 4: even error paths come with exact responses ('run the command in `details.command`'). | 5 / 5 |
Workflow Clarity | Six clearly sequenced sections (Discover → MCP facts → Configure → Wire → Verify → Report) with explicit validation checkpoints: 'Dry-run first: it sends nothing and lists every missing setting. If anything is missing, stop', atomic batch semantics for 'intents add', a six-item Verify section including a negative match test and a failure-mode test ('Unset `IRIS_INTENT_ROUTER_API_KEY` and confirm `ConfigurationError` names it'), and feedback loops for error recovery ('Wrong intent first: add phrasings to the right intent or raise its `minScore`, re-run `intents add`'). Not 4: validation is explicit and feedback loops cover both tuning directions, matching the anchor-5 example's structure. | 5 / 5 |
Progressive Disclosure | No bundle files exist (no references/, scripts/, or assets/ directories), and the body is a single ~136-line file with clear section headers, a consistent question/command/table pattern, and an external pointer to 'iris intent-router <command> --help' as the full reference. Structure is good and nothing is buried or nested. Not 5: the under-50-line simple-skill exception does not apply at this length, and portions of the CLI operational rules (error handling per error code, key minting workflow) are dense enough that they could live in a one-level reference file, keeping SKILL.md as a tighter overview. Not 3: there is no misplacement — inline content is appropriate to the workflow and navigation is easy. | 4 / 5 |
Total | 19 / 20 Passed |