Content
88%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.
A well-structured, highly actionable skill body: a six-phase workflow with real validation checkpoints, concrete commands and templates, and proper delegation of filing rules to external files. Its weaknesses are minor — redundant restatements of the notability gate and fetch-fail rules, and slightly loose placement of the back-link convention.
Suggestions
Consolidate the fetch-fail guidance into one section (either "Edge Cases" or "When it fails") and cross-reference from the other to remove duplication.
Fold the back-link format paragraph into the Contract section (or the referenced conventions file) so the Returns block is not followed by a stray rule.
State the notability gate once in Phase 3 and reference it from the Contract and Anti-Patterns instead of restating the full condition three times.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and assumes Claude's competence with no concept over-explanation, but the notability-gate rule and the inline author-credit format are each stated three times (Contract, Phase 3, Anti-Patterns), and the fetch-fail case is described in both "Edge Cases" and "When it fails". This is efficient with minor trimmable repetition (anchor 4) rather than noticeably verbose (anchor 3). | 4 / 5 |
Actionability | Concrete commands ("gbrain files upload-raw <file> --page <slug>", "gbrain sync", "gbrain write-request <request_id>"), an exact markdown page template, a precise back-link format, and a typed return contract (page_path, author_path, cross_links, status) give copy-paste-ready guidance. For an instruction-only skill this fully matches the anchor-5 executable-guidance standard. | 5 / 5 |
Workflow Clarity | Six clearly sequenced phases (fetch, upload raw, identify author, save, analyze, sync) with explicit validation checkpoints (pre-ingest duplicate-URL search, notability gate, write_pending polling) and a dedicated error-recovery section. The destructive/batch cap does not apply because verification steps are present; this matches the anchor-5 feedback-loop example. | 5 / 5 |
Progressive Disclosure | Sections are well-organized (Contract, Phases, Output Format, Edge Cases, When it fails, Anti-Patterns) with one-level, well-signaled external references ("skills/_brain-filing-rules.md", "skills/conventions/quality.md", the agent operator protocol). Minor organization gaps remain: the fetch-fail rule is duplicated across two sections, and the back-link format paragraph sits loosely after the Returns block. This fits anchor 4 (good structure, minor gaps) rather than anchor 5's easy-navigation standard. | 4 / 5 |
Total | 18 / 20 Passed |