Content
81%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-engineered, information-dense skill body: a clearly sequenced 7-step workflow with explicit error-recovery paths and a scoring gate, concrete parsing and generation rules, and a clean one-level-deep bundle split. The main defects are the duplicated error-handling content between Step 2 and the Error Handling table, unsourced {DOMAIN}/{OWNER}/{REPO} URL placeholders, and two referenced example files (examples/version-output.md, examples/week-output.md) that are absent from the bundle.
Suggestions
Remove the duplicate error responses: keep them only in the Error Handling table (or only inline in Step 2) instead of repeating the same four error/response pairs in both places.
Add a one-line instruction on where to resolve the URL placeholders (e.g., derive {OWNER}/{REPO} from `git remote get-url origin` and {DOMAIN} from the project config or landing-page convention) so generated posts are copy-paste ready.
Fix the progressive-disclosure gap: either add the missing `examples/version-output.md` and `examples/week-output.md` files to the bundle or drop the two dead pointers from the Reference Files section.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and table-driven with essentially zero explanation of concepts Claude already knows — parsing rules, scoring formula, and flags are all given as compact tables. The one real redundancy is that error responses appear twice (inline in Step 2 — "**Error: version not found** -> List last 5 versions, suggest `latest`" — and again, nearly verbatim, in the Error Handling table), which fits anchor 4 (efficient; minor instances that could be trimmed) rather than anchor 5's 'every token earns its place'. | 4 / 5 |
Actionability | Highly concrete: exact input modes, a fully specified scoring formula ("score = (category_weight * 3) + (has_number * 2) + ..."), per-format link targets, output file naming patterns, and a complete flag table. Two minor gaps keep it below anchor 5: the URL placeholders (`https://{DOMAIN}/`, `https://github.com/{OWNER}/{REPO}`) are never sourced (no instruction on where to discover the domain or owner/repo), and the Step 7 output example uses a broken nested-code-fence workaround (`` `` ` ``) that renders incorrectly if copied. | 4 / 5 |
Workflow Clarity | The 7-step workflow (with a clearly-specified Step 4b insertion point for --interactive) is explicitly sequenced, includes a validation gate ("If all scores < 3: Output 'No social content recommended...' and exit (unless `--force`)"), and provides feedback loops for error recovery across six enumerated failure modes in the Error Handling table — directly matching anchor 5's 'clear sequence with explicit validation steps; feedback loops for error recovery'. Anchor 4 would require missing checkpoints, which are present here. | 5 / 5 |
Progressive Disclosure | Good structure overall: SKILL.md stays an overview, with a dedicated Reference Files section describing all nine bundle files, and the actual bundle (3 references/ + 4 assets/) is flat, one level deep, with no nested reference chains. However, two of the referenced paths — `examples/version-output.md` and `examples/week-output.md` — do not exist in the bundle, so those pointers are dead. This is a navigation defect beyond anchor 4's 'minor organization gaps' in spirit, but the references are clearly signaled and appropriately split, placing it above anchor 3; it cannot be anchor 5 ('easy navigation') while two listed files are missing. | 4 / 5 |
Total | 17 / 20 Passed |