Content
80%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
A code-heavy, actionable skill body that efficiently teaches Remotion-specific conventions without padding. It would benefit from explicit inline validation/feedback loops in the rendering workflows and from splitting reference-like detail into bundle files.
Suggestions
Add an explicit validate-fix-retry loop to the Manual Rendering workflow (e.g., render fails -> check Error Codes table -> fix props/config -> re-render) instead of deferring recovery to a separate section.
Move the Props Schema, Common Patterns, and Error Codes into reference files under references/ and link to them one level deep, keeping SKILL.md a lean overview.
Add a validation checkpoint in the rendering flow (e.g., run `npx tsc --noEmit` before render, verify output .mp4 exists after) to strengthen workflow_clarity.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body avoids explaining what Remotion or libraries are; it is code- and rule-dense with app-specific conventions (frame-based animation, determinism, props schema, error codes) that Claude does not already know, so every section earns its place. | 3 / 3 |
Actionability | Executable TypeScript (useCurrentFrame, interpolate with clamping, spring), a typed props interface, and copy-paste render commands provide concrete, complete guidance rather than vague direction. | 3 / 3 |
Workflow Clarity | The Auto-Generation Flow lists a clear sequence with an explicit PASS checkpoint, but the Manual Rendering workflow has no inline validation steps and error recovery is deferred to a separate Error Codes table rather than expressed as validate-fix-retry feedback loops. | 2 / 3 |
Progressive Disclosure | Sections are well organized, but the ~200-line body inlines reference-like material (Common Patterns, Props Schema, Error Codes) that could live in separate files, and the Reference section points to paths outside the bundle rather than splitting detail one level deep. | 2 / 3 |
Total | 10 / 12 Passed |