Content
81%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.
The body is an exceptionally concrete routing and execution guide: executable JSON call shapes, exact field-placement and pagination rules, and a validated multi-step workflow with explicit error-recovery loops. Its weaknesses are length and drift — several sections cover host/WebAgent behavior and product meta-details rather than the API surface — and references to category files that could not be verified in the bundle.
Suggestions
Trim or relocate sections unrelated to the MoviePilot API surface — e.g., the WebAgent choice-callback/snapshot internals in "Agent Task And Conversation State" and the memory/skill tooling guidance in "Reusable learning" — to cut token cost without losing routing value.
Move route-inventory meta-details (widget Bearer semantics, image proxy maxWidth/quality contract, the app/agent/policy/api.py drift test) into the linked docs files and keep only the routing-relevant boundary statements inline.
Standardize reference locations (api/, docs/, root-level code-execution.md, ../../docs/mcp-api.md) under one consistent bundle layout and ship the referenced category files with the skill so every link resolves.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The core routing and contract prose is dense with unique operational rules, but sections drift off-skill: "Agent Task And Conversation State" (WebAgent choice callbacks, snapshot replacement, idle workers), "Reusable learning" (memory tool guidance), and meta paragraphs like "The Web image proxy's maxWidth and quality parameters" and "The maintained route-by-route inventory is the fixed registry in app/agent/policy/api.py". That is more than the "minor instances of over-explanation" of the 4 anchor, though the text is never padded, so it stays above 2. | 3 / 5 |
Actionability | Quotes: the executable call shapes {"operation_id":"media.search","path_params":{},"query":{"title":"The Wandering Earth","type":"media"},"body":{}} and {"operation_id":"subscription.execution.list","query":{"limit":10}}, the pagination recipe "send query={"page":1,"count":1} and read collection.total_count", and the verification-contracts table with exact path_params/query fields. Guidance is copy-paste ready and covers the common read/download cases, matching the 5 anchor. | 5 / 5 |
Workflow Clarity | The "Overall Workflow" gives a 7-step sequence with explicit validation checkpoints and feedback loops: "Inspect success, execution_outcome, errors, empty results… before reporting", "An invalid_input response names the failing field… correct that field and all required fields before retrying", and "When a write returns unknown, never retry it; verify the actual state with a supported read operation first". Destructive/batch operations (storage.delete, transfer.history.redo_batch, database.backups.delete) are covered by these checkpoints and read-back verification, so the cap-at-3 rule does not apply. | 5 / 5 |
Progressive Disclosure | Quotes: "Detailed operation contracts live in the linked category files under api/; load only the one category file needed" and the 22-row "API Category Index" table with namespaces and use-for descriptions — a well-signaled, one-level-deep reference design. However, none of the referenced files (api/<category>.md, docs/widgets.md, code-execution.md, ../../docs/mcp-api.md) exist in the reviewed bundle, and reference locations are inconsistent (api/, docs/, skill root, two levels up), which are minor organization gaps versus the fully verified split of the 5 anchor. | 4 / 5 |
Total | 17 / 20 Passed |