Content
85%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.
The body is an exceptionally lean operational contract: every sentence instructs, dense phase headers sequence the work end to end, and validation gates plus repair loops are explicit at every approval, paid batch and completion. Actionability is high — named server actions with exact fields and thresholds — though nothing is copy-paste executable since exact schemas are deliberately deferred to live inspection. The one structural defect is a dangling, unclearly-signaled reference to references/recipe-format.md, which does not exist in the delivered bundle.
Suggestions
Ship the cited references/recipe-format.md file in the bundle, or remove the citation and state the recipe format inline — as delivered, the skill's only self-referenced path is unresolvable.
Turn the recipe-format citation into a clearly signaled link with a one-phrase description of its contents (e.g. '**Rebuild recipe**: See [recipe-format.md](references/recipe-format.md) for the exact recipe schema to store') instead of a bare parenthetical.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Every sentence is an instruction ("Custom skips styles. Ask which product or service to advertise; wait even with one saved."; "Never infer consent or approve yourself. Honor Stop.") — there is no explanation of concepts Claude already knows and no padding to trim, matching 'lean and efficient; assumes Claude's competence; every token earns its place'. Not 4 because no instance of over-explanation exists; the two-line owner-facing preamble is minimal and functional. | 5 / 5 |
Actionability | Guidance names concrete operations with exact fields and thresholds ("call video_approve with the custom phase, review_token and total_credits from custom_review.approval_quote ... expected_revision and their words as user_quote"; "run video_read with include [\"blockers\"] (free)"; "up to 20 per call"), which is concrete command-level guidance with minor gaps — matching anchor 4. Not 5 because nothing is copy-paste executable: exact call syntax is deferred to live schema inspection ('inspect live schemas before writes'), a flexibility that is explicitly justified but still leaves the literal command forms unstated; not 3 because this is far more specific than pseudocode. | 4 / 5 |
Workflow Clarity | The body sequences the work in ordered phase sections (Load and resume → Reference, product and creative contract → Save and review twice → Produce, assemble and deliver) with pervasive explicit validation gates ("Before each approval, paid batch or completion, run video_read with include [\"blockers\"] ... and fix every blocking and must-fix finding in one pass"; "only checked bytes enter the final"), feedback loops ("Reject double mouths, extra hands, smears and garbled labels ... Re-make affected sources within approval"), and a checklist (the 8-item Non-negotiables). This matches anchor 5; not 4 because checkpoints gate every approval, paid batch and completion rather than only 'most' being present. | 5 / 5 |
Progressive Disclosure | The bundle ships no references/, scripts/ or assets/ directories, yet the body cites "references/recipe-format.md" ("Store the rebuild recipe (references/recipe-format.md)") as a bare parenthetical — a reference that is neither clearly signaled nor resolvable in the delivered bundle, and the only other file pointer ("harness references/talking-actor.md") belongs to a different skill. This matches 'references present but not clearly signaled'; not 4 because the unresolvable self-reference is more than a minor organization gap, not 2 because the ~46-line body is well-sectioned and its content appropriately lives inline. | 3 / 5 |
Total | 17 / 20 Passed |