Content
90%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 lean, highly actionable skill file: exact MCP/CLI invocations, literal output formats, and explicit reply-parsing rules leave little ambiguity about what to do in each flow. The only notable weaknesses are the missing index-validation before the destructive `set_defaults` batch replacement and a single-file structure that is slightly heavier than the ideal overview-only SKILL.md.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with skill-specific operational facts Claude could not know (e.g. "Only `scan` walks the disk", "at most two directory levels deep", "text between tool calls may not render, and dialog previews truncate long lists") and contains essentially no explanation of known concepts. Every section earns its tokens; no padded passages found. | 5 / 5 |
Actionability | It gives copy-paste-ready tool calls with exact arguments ("{ \"action\": \"query\" }") and exact CLI commands ("uvx --python '>=3.12' --from ouroboros-ai ouroboros detect [path]"), plus literal response strings to display ("No git repositories or worktrees found.") and concrete reply-parsing rules covering normal, keep/none, and invalid-input cases. | 5 / 5 |
Workflow Clarity | The default flow is a clear 3-step sequence with decision points ("If `meta.total` is 0 ... run the scan below") and an error-recovery loop ("anything else → ask again in plain text, never guess"), and `detect` has a failure path ("could not propose any verifiable commands"). Minor gap: `set_defaults` is a destructive batch replacement ("it replaces every existing default") with no explicit pre-call check that the replied numbers map to registered repos, though it does warn and confirm afterward. | 4 / 5 |
Progressive Disclosure | No bundle files exist and none are referenced; the body is well-sectioned (Loading MCP tools, Default flow steps, Subcommands, Footer) and stays a single level deep. It is ~100 lines of inline detail (scan boundaries, find ranking rules), slightly beyond the under-50-line single-file ideal, so a small amount of content (e.g. scan-boundary or detect specifics) could arguably live in reference files. | 4 / 5 |
Total | 18 / 20 Passed |