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.
An exemplary skill body: a gated 5-step workflow with quantified validation and error-recovery loops, fully executable commands, precise output-format specs, and textbook progressive disclosure via an API-map index. The only weaknesses are mild redundancy in the three separate scope/boundary sections and a date-stamped baseline table that will age.
Suggestions
Consolidate the scope guidance: the front-door blockquote, 'When to use', and 'What this skill explicitly does NOT do' repeat the same boundaries; merge the latter two and drop the trigger-phrase list that duplicates the frontmatter description.
Show the exact lint invocation in Step 1 (e.g. `python scripts/lint_source.py <src-dir>`) the way Steps 2 and 4 show theirs, so every step's first command is copy-paste ready.
Move the date-stamped validated-baseline table into references/eval.md next to the thresholds it documents, keeping the body's SSIM guidance to the rule rather than the snapshot numbers.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and assumes Claude's competence — no concept explanations, no library tutorials — and every workflow step carries a command, rule, or spec. Minor trimming is possible: scope boundaries are stated three times (front-door blockquote, "When to use", "What this skill explicitly does NOT do"), largely restating the frontmatter, and the date-stamped validated-baseline table ("as of 2026-04-27") is time-sensitive detail. Anchor 4 ('efficient; minor instances that could be trimmed') fits; anchor 3's 'unnecessary explanation' is too harsh since nothing teaches known concepts. | 4 / 5 |
Actionability | Fully executable guidance throughout: lint via scripts/lint_source.py with named blockers/warnings/info, catalog search with the exact `npx hyperframes catalog --query ... --json` invocation, a precise Step-3 output spec (data attributes, paused gsap.timeline, window.__timelines registration), and copy-paste render/diff commands (`npx remotion render <CompositionId> out/baseline.mp4`, `npx hyperframes render --skill=... --output ...`, render_diff.sh) with the pixel-format fix (Config.setVideoImageFormat("png") + Config.setColorSpace("bt709")). Matches anchor 5; the only soft spot is that the lint script's exact CLI arguments are implied rather than shown. | 5 / 5 |
Workflow Clarity | A clear 5-step sequence (lint → plan → generate → validate → document) with explicit checkpoints and feedback loops: blockers stop translation and route to escape-hatch.md; a failed SSIM diff triggers frame_strip.sh to localize diverging frames before re-reading the relevant reference; thresholds are quantified per tier. This matches anchor 5 (explicit validation steps with error-recovery loops); the batch/destructive cap at 3 does not apply because validation is central. | 5 / 5 |
Progressive Disclosure | Model progressive disclosure: the SKILL.md is an overview; api-map.md is an index with a source-construct → reference table covering all 11 reference files (all verified present in the bundle), scripts and assets are referenced by path and exist, and it explicitly instructs "Don't load all of them — load only what the specific source needs." References are one level deep. One minor blemish: the conditional link to ../hyperframes/references/plugin-installation.md does not resolve in this standalone checkout, but it is explicitly scoped to plugin installs, so navigation within the bundle remains clear. | 5 / 5 |
Total | 19 / 20 Passed |