Content
92%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 excellent operational skill body: fully executable commands, a tightly sequenced five-step workflow saturated with validation checkpoints and feedback loops, and clean separation between orchestration instructions and script implementation. The only weakness is mild redundancy in repeated anti-bypass/security warnings that could be consolidated for token efficiency.
Suggestions
Consolidate the repeated anti-bypass warnings (sections 3, 4, and 5 each restate "do not bypass checks") into a single rule in 核心约束 and reference it once, trimming token cost.
Consider moving the detailed MoviePilot token/Device-Flow permission rules (core-constraints bullets on scopes and PAT permissions) into a short referenced file, keeping only the operating rules inline in SKILL.md.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and prescriptive, teaching nothing Claude already knows (no explanations of Git, forks, or PR concepts), but the anti-bypass warning recurs about three times ("不要绕过检查", "不要绕过校验", "禁止通过 --force ... 绕过检查") and parts of the 核心约束 section could be consolidated — minor instances that could be trimmed. | 4 / 5 |
Actionability | Fully executable throughout: complete bash invocations with arguments for all four scripts, a concrete draft JSON template with field-by-field explanation, and a verbatim confirmation-gate prompt — copy-paste ready guidance covering the common path. | 5 / 5 |
Workflow Clarity | Five numbered steps with explicit validation checkpoints at every stage: outputs that must be saved and re-read ("必须用 read_file 读取 preview_file 全文"), a hard confirmation gate before submit, submit-time re-verification (token user, base SHA, origin remote, branch and file-set consistency, remote SHA match), and concrete failure/retry handling with payload reuse. Despite pushing to a remote, validation is abundant, so no cap applies. | 5 / 5 |
Progressive Disclosure | SKILL.md stays a workflow overview while implementation lives in scripts/: four clearly signaled one-level-deep script paths in code blocks, all verified to exist in the bundle (clone_repository.py, collect_pull_request.py, prepare_pull_request.py, submit_pull_request.py), with no nested references and no bulk detail inlined that belongs in a separate file. | 5 / 5 |
Total | 19 / 20 Passed |