Content
77%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 delivers highly actionable, well-sequenced workflows with genuine validation and error-recovery loops — the strongest part of the skill. Its weaknesses are repetition and host-implementation detail that inflate token cost, and a monolithic single-file layout that inlines reference-grade material instead of splitting it out.
Suggestions
Deduplicate the session-bound captcha guidance (stated in the `view_image` tool bullet, captcha step 2, and step 4) into a single authoritative statement and trim host implementation details (60-second scan cadence, TaskRegistry rollback mechanics) to the one behavioral rule the agent needs: re-open and re-verify after long waits, close sessions explicitly.
Move the detailed MoviePilot site workflows — especially the graphic-captcha login ladder and the session idle policy — into a references/ file, keeping SKILL.md as a concise overview with one-level-deep pointers per the progressive-disclosure model.
Add Chinese trigger phrases (e.g., 打开网页, 截图, 登录失效, 更新Cookie) to the description and When To Use section, since the skill's own Examples show its users prompting in Chinese.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly efficient, command-centric, and free of explanations of concepts Claude already knows, but it needs tightening beyond "minor" trimming: session-bound captcha guidance is repeated across the Tools section and steps 2 and 4 of the captcha workflow, and host implementation details ("宿主在正常及安全模式下每 60 秒扫描一次", TaskRegistry scan/rollback mechanics) exceed what the agent needs to act on. It is well above 2, since there is no padded library-tutorial prose, but the repetition and implementation detail keep it below 4. | 3 / 5 |
Actionability | Nearly every instruction is a copy-paste-ready command — `browse_webpage action="goto" url="https://example.com"`, `click_ref`/`fill_ref`/`select_ref` usage, an executable `evaluate` script, and full `moviepilot_api operation_id=...` calls — and the Examples section covers the common cases end-to-end. This matches the 5 anchor and exceeds 4 because placeholders only appear where values are inherently user-supplied (credentials, refs). | 5 / 5 |
Workflow Clarity | Multi-step processes are explicitly sequenced with validation checkpoints and feedback loops: the core "navigate → observe → one small action → verify" loop, the site diagnosis workflow that re-runs `site.test` to confirm the fix, and the captcha workflow's recovery path (OCR fail → visual recognition → refresh once → manual input) with a hard three-attempt limit and post-rejection re-inspection. This matches the 5 anchor; 4 would require some checkpoints to be merely implied, which they are not. | 5 / 5 |
Progressive Disclosure | Section headers are clear and flat, but the entire skill is a single ~310-line file with no bundle files, and roughly 120 lines of specialized detail (the captcha login edge-case ladder, the session idle/TaskRegistry lifecycle policy) are inlined where the rubric expects them in one-level-deep reference files. This fits the 3 anchor ("content that should be separate is inline") better than 4 — the gap is more than minor — while remaining above 2 since structure and navigation within the file are good. | 3 / 5 |
Total | 16 / 20 Passed |