Content
77%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 exceptionally actionable with a well-sequenced, validation-heavy workflow and a concrete completion condition. Its weaknesses are token efficiency — generic advice, an obvious folder table, and scattered version numbers inflate a ~550-line body — and progressive disclosure, since large reference-grade sections are inlined with no bundle files to absorb them.
Suggestions
Move reference-grade detail (the limits and refresh-safety cap tables, embedder resolution order, Jikji store internals, and the full Hister setup procedure) into bundle files under references/, keeping SKILL.md as a concise overview with clearly signaled one-level-deep links.
Trim guidance Claude can derive itself, such as the model-selection paragraph ("reliable tool calling and structured output... high output TPS, and low first-token latency") and the OS-specific recommended document-folders table.
Consolidate scattered version requirements (mailcrawl 0.2.2 vs 0.1.3, lazykatok 0.3.3, MinSync 0.4.12+) into a single minimum-versions or deprecated-versions section so time-sensitive details are localized.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Most content is tool-specific fact Claude cannot know (config keys, defaults, behaviors), so it is not a 2, but the ~550-line body includes trimmable material: generic model-selection advice, an obvious OS document-folders table, and many inline version numbers ("0.2.2", "0.1.3 or earlier", "lazykatok 0.3.3", "MinSync 0.4.12+") outside any deprecation section. Fits "mostly efficient but could be tightened" rather than 4's "minor instances". | 3 / 5 |
Actionability | Fully executable throughout: copy-paste commands ("autorag init --search-paths ... --model-provider PROVIDER --model-id MODEL", "autorag setup --format json", "brew services start hister", "command -v dupey || cargo install dupey --locked"), complete config JSON blocks, and a field/default/controls table cover the common cases, matching the 5 anchor. | 5 / 5 |
Workflow Clarity | Clear sequence (install → inspect → configure → approve roots → init → datasources → verify → MCP → schedule) with explicit validation: "Configuration alone is not a successful setup" backed by status/health/refresh/search verification commands, probe-first datasource setup with skipped-list reporting, Hister's numbered steps with "hister doctor" and restartRequired/reindexRequired checks, and a Completion condition checklist. Destructive operations (refresh, --force) are guarded. Matches the 5 anchor. | 5 / 5 |
Progressive Disclosure | Section headers keep the body navigable (ruling out 2's "no section headers" example), but reference-grade detail — the retrieval-caps table, embedder resolution order, Jikji store internals, and the full Hister procedure — is inlined rather than split into bundle files, and references (docs/minsync-setup.md, docs/datasource-skills.md) are mentioned parenthetically mid-text and are absent from the bundle. Fits "some structure; content that should be separate is inline"; not 4 because the split-out opportunity is extensive rather than minor. | 3 / 5 |
Total | 16 / 20 Passed |