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.
This is a strongly engineered skill body: checklist-driven workflow with decision tables and validation gates, fully executable commands with flag-level precision, verified one-level-deep reference files, and disciplined table-based conciseness that assumes Claude's competence. The only weakness is mild: the remote-api architecture prose and the feature-comparison table could be tightened or delegated to references.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and table-driven with essentially no explanation of concepts Claude already knows — e.g., the troubleshooting table, the EXTEND.md key table, and the one-line 'Never pre-convert markdown to HTML' rule all carry pure operational signal. It falls short of 5 only in a few spots that could be trimmed or moved to references: the multi-sentence prose in Step 2 explaining the remote-api SSH SOCKS5 architecture ('Rendering, image processing, draft assembly and HTML rewriting stay local; only the HTTPS calls ... go through an SSH SOCKS5 forward') and the 13-row Feature Comparison table partially duplicate information already conveyed by the Step 2 method table. Not 3: these are minor instances of over-explanation in an otherwise efficient document, not a pattern of unnecessary explanation. | 4 / 5 |
Actionability | Guidance is copy-paste ready throughout: full bash invocations with every flag ('${BUN_X} {baseDir}/scripts/wechat-api.ts <file> --theme <theme> --remote [--remote-host <host>] ...'), the exact API endpoint and payload rules ('POST https://api.weixin.qq.com/cgi-bin/draft/add?access_token=ACCESS_TOKEN', 'article_type: news (default) or newspic'), concrete fallback chains ('CLI --cover → frontmatter (coverImage / featureImage / cover_image / cover / image) → imgs/cover.png → first inline image → stop and request one'), and specific plain-text file-naming rules ('post-to-wechat/YYYY-MM-DD/<slug>.md'). Not 4: commands are executable as written and the common cases (api, remote-api, browser, image-text) each have a complete example with no gaps. | 5 / 5 |
Workflow Clarity | The Article Posting Workflow is an explicit checklist ('- [ ] Step 0: Load preferences ... - [ ] Step 5: Report completion') with per-step decision tables (input-type detection, method selection, metadata field resolution), an explicit validation gate in Step 0 ('if missing, finish first-time setup before any other question'), a pre-flight permission check with a failure→fix table, and a Troubleshooting section with feedback loops for error recovery (e.g., 'errcode 40164' → add egress IP to the allowlist, SOCKS proxy not ready → check key/host/timeout). Not 4: checkpoints are explicit at nearly every step, including the mandatory cover-image stop condition and the completion report enumerating what must be verified. | 5 / 5 |
Progressive Disclosure | SKILL.md is a clear overview that delegates details to well-signaled, one-level-deep references that all exist in the bundle: 'references/config/first-time-setup.md', 'references/multi-account.md', 'references/api-setup.md', 'references/image-text-posting.md', 'references/article-posting.md' — verified present in ./references/ (plus ./references/config/), and none of these files point to further nested references. The Script Directory table gives a purpose for each script, matching the actual scripts/ contents. Not 4: every reference is clearly signaled at its point of use, the split is appropriate (details out, decisions in), and navigation requires no second hop. | 5 / 5 |
Total | 19 / 20 Passed |