Content
65%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
A highly actionable, well-organized reference with concrete executable code and commands, but it is a long monolithic file with no progressive disclosure to bundle files and carries redundant content (notably the gotchas section restating earlier guidance). Workflow sequencing is present but lacks explicit validation feedback loops for destructive push operations.
Suggestions
Move the duplicated gotchas into a single authoritative location (or trim them to non-obvious pitfalls only) to remove verbatim repetition of the snake_case, featureId, and interval rules already covered above.
Extract the detailed SDK/CLI reference and the project-file map into a reference bundle file (e.g. references/autumn-reference.md) linked from SKILL.md, so the main file stays a lean overview with one-level-deep navigation.
Add an explicit validate→fix→retry loop around `atmn push --prod` (e.g. run `atmn preview`, review the diff, fix config, re-preview, then push) to raise workflow clarity for the destructive production-push path.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly efficient domain-specific guidance Claude would not already know, but the ~350-line length includes redundancy — the 8-item 'Common Gotchas' section repeats earlier content verbatim (e.g. featureId always 'ai_usage', snake_case IDs, reset/price interval exclusivity) and the 'Upstream Grounding' paragraph is wordy, so it could be tightened. | 2 / 3 |
Actionability | It provides fully executable TypeScript and bash examples, concrete file paths, and specific function signatures (e.g. `autumn.check({ ... requiredBalance: credits, sendEvent: true })`, `bun x atmn push --prod`), matching the copy-paste-ready anchor. | 3 / 3 |
Workflow Clarity | Sequences exist (the `atmn preview` → `push` → `push --prod` flow and the getOrCreate-then-check flow) with a dry-run checkpoint, but the skill is structured as a reference rather than a sequenced workflow, and risky/destructive operations like `push --prod` lack an explicit validate→fix→retry feedback loop, capping it at 2. | 2 / 3 |
Progressive Disclosure | Sections and tables are well-organized, but no local bundle files exist and all detailed material (full API patterns, the gotchas list, the project-file map) is inline in a single ~350-line SKILL.md rather than split into one-level-deep referenced files, matching the 'content that should be separate is inline' anchor. | 2 / 3 |
Total | 9 / 12 Passed |