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.
This is a dense, highly executable operational runbook: every step has exact commands and selectors, validation checkpoints and retry loops are built in throughout, and pitfalls document real failure signatures. The weaknesses are structural — the same constraints are repeated across sections, inflating token cost, and the whole thing is one monolithic file instead of an overview with split-out references.
Suggestions
Deduplicate rules stated three times (cookie/UA closure, fail-first queue, ~300字 guidance): state each once in 必做约束/Session 共享约束 and reference it from other sections.
Split the login/login-confirm QR protocol and the Pitfalls catalog into references/ files (e.g. LOGIN.md, PITFALLS.md), keeping SKILL.md a lean overview with clear one-level-deep pointers.
Move the dated daemon note (「2026-08-22 起已默认关闭 idle 自退」) into a deprecated/old-patterns note so time-sensitive history does not pad the active workflow.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Information density is high and nothing explains concepts Claude already knows, but several rules are restated verbatim three times across sections — 「不导出 cookie/UA」 (前置条件 twice, then Session 共享约束), the fail-first queue (intro, 必做约束, Session 共享约束), and 「约 300 字为建议值」 (输入, Step 6, 入库衔接约束) — plus a dated note (「2026-08-22 起已默认关闭 idle 自退」) sitting inline in a pitfall rather than a deprecated section. The duplication goes beyond the 'minor instances' of anchor 4, so 3 is the best fit. | 3 / 5 |
Actionability | Every step ships copy-paste-ready commands: exact invocations like `camoufox-cli --session wechat-channel --persistent --json open "https://channels.weixin.qq.com/platform/post/create"`, complete eval JS snippets including the mousedown/mouseup/click dispatch walk, scoped selectors such as `input[type=file][accept*="image"]`, the `wechat-channels-publish fill --body @/绝对路径/body.txt` wrapper, and literal error strings to match (「session wechat-channel 正忙」). Covers common cases including fallbacks — matches the fully-executable anchor. | 5 / 5 |
Workflow Clarity | The 8-step flow is clearly sequenced with explicit validation checkpoints and feedback loops: transcode polling via `snapshot -s` state classes with a 3-minute cap and failure signatures, short-title verification by eval-reading `.value` with a re-fill retry loop (「为空 → 重新 fill 再校验」), a disabled-button check before publish, and post-publish confirmation via URL/redirect, backed by an error-handling table with recovery actions. This matches the explicit-validation-with-error-recovery anchor. | 5 / 5 |
Progressive Disclosure | No bundle files exist (no references/, scripts/, or assets/), so all ~290 lines live in one SKILL.md. Section headers are good, but self-contained blocks — the login/login-confirm QR protocol, the full Pitfalls catalog, and 入库衔接约束 — are inlined where they clearly belong in separate reference files. Fits 'some structure but content that should be separate is inline'; not 4 because there are no one-level-deep reference files at all for a document of this size. | 3 / 5 |
Total | 16 / 20 Passed |