Content
67%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 gate skill: concrete CLI commands, a sequenced decision protocol with a validation checklist, and explicit error paths for ambiguity and missing types. Its main weakness is redundancy — the no-hardcoding principle is repeated across four sections and the Critical section restates the Contract — plus an ironic hardcoded book taxonomy that contradicts the skill's own single-source-of-truth rule.
Suggestions
State the 'no hardcoded directory table' rule once (e.g., in Hard Rules) and remove the duplicated restatements in the Contract, the Critical section, and Anti-Patterns — the Critical section can be deleted outright since it adds nothing the Contract does not already guarantee.
Remove the hardcoded 11-category book list from Step 3 or replace it with a pointer to the pack ('walk `page_types[]` for the media/books branch'), since it violates the skill's own no-inline-taxonomy principle and will drift from the active pack.
Add a brief recovery loop for Step 5 checklist failures (e.g., 'if any box fails, return to Step 1/2 and re-derive the path before writing') to complement the existing missing-type error path, and add one worked example of filing a single page to ground the abstract 'prefer the more specific type' guidance.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly dense and project-specific, but the 'no hardcoded directory table' rule is stated four separate times ('This skill guarantees', the 'Critical' section, Anti-Patterns, and Hard Rules), and the Critical section largely restates the Contract. It also hardcodes an 11-item book category list ('psychology, philosophy, spirituality, business, media-and-divorce...') in Step 3 despite the skill's own single-source-of-truth principle, which is inline data that could be dropped or referenced. This matches 'mostly efficient but includes some unnecessary explanation or could be tightened' — not 2, since no section explains concepts Claude already knows, and not 4, since the duplication is systematic rather than minor. | 3 / 5 |
Actionability | Guidance is largely executable and copy-paste ready: 'gbrain schema show --json | jq '.page_types[] | select(.primitive == "entity")'', 'gbrain schema review-orphans --json', 'gbrain doctor --json | jq ...', plus a concrete 7-question decision walk and a filled-in output template. It falls short of 5 only because a few directives stay abstract — 'prefer the more specific one (the one with the more specific path prefix)' offers no tie-break procedure, and there is no worked end-to-end example (e.g., filing one actual book). Well above 3: no pseudocode, all commands runnable. | 4 / 5 |
Workflow Clarity | The 5-step Decision Protocol is clearly sequenced with an explicit validation checkpoint — Step 5 'Validate before writing' is a 4-item checklist — and explicit error paths (ambiguity surfaces '2 candidates via `skills/ask-user/`', no-match type signals EIIRP Phase 3). This is well past anchor 3 (checkpoints present). It sits below anchor 5 because the recovery loop for generic checklist failures is unspecified — only the missing-type failure has a defined fix-and-retry path — and the drift-detection flow has no validation step. | 4 / 5 |
Progressive Disclosure | The single-file body (~175 lines) is well-sectioned with clear headers (Purpose, Contract, Decision Protocol, Integration, Drift Detection, Output Format, Anti-Patterns, Hard Rules), making navigation easy, and no nested or buried references exist. This matches 'good structure; most content appropriately placed; minor organization gaps'. Not 5: the skill exceeds 50 lines and inlines content that could be split out (the book category enumeration, the changelog, the cross-skill integration list), though as a single-file skill with no bundle directory this is a minor gap rather than the inlining of anchor 2/3. | 4 / 5 |
Total | 15 / 20 Passed |