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 — exact commands, complete config examples, pinned versions, explicit validation and error-recovery paths — and the workflow culminates in a genuine completion checklist. Its weaknesses are token weight (much inline detail a given setup never needs) and absent progressive disclosure: everything lives in one monolithic SKILL.md with no reference files.
Suggestions
Move the deep-dive sections (record-cut mechanics and benchmark, Hister setup, embedder resolution order, limits tables) into reference files under references/ and keep a short summary plus a clearly signaled link in SKILL.md.
Deduplicate the credential-safety guidance (stated ~5 times across Safety, model config, embedder, Jev, and star-repo sections) into one authoritative statement.
Trim operator-facing detail that does not change agent behavior, such as the maintainer's nDCG/MRR benchmark figures and the Jikji 0.1.x migration history.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Almost all content is tool-specific fact Claude cannot know (config keys, defaults, error codes), but at ~640 lines there is real padding: maintainer benchmark figures ("nDCG@10 0.646 → 0.656, MRR@10 0.588 → 0.600..."), credential-handling warnings repeated several times, and long prose passages that could be tightened into tables. It fits "mostly efficient but could be tightened" better than the noticeably-verbose anchor below because generic concept explanation is nearly absent. | 3 / 5 |
Actionability | The body gives copy-paste-ready commands ("bun install -g @autorag/librarian", "autorag init --search-paths ... --model-provider PROVIDER --model-id MODEL", "gh api -X PUT /user/starred/..."), complete config JSON blocks for every scenario, pinned versions ("jikji-cli 0.2.1", "@nomadamas/mailcrawl@0.3.0"), and per-error-code remediation ([decision-model-unavailable] reason table). Fully executable guidance covering the common cases — the top anchor. | 5 / 5 |
Workflow Clarity | A clear install → inspect → configure → approve roots → init → verify → MCP → schedule sequence is capped by an explicit "Completion condition" checklist; validation is concrete (autorag health --json, one real structured search, MCP tools/list plus known-phrase search) and error recovery loops exist (per-reason decision-model remediation, restartRequired/reindexRequired handling, mass-delete suppression). This matches the anchor with explicit validation, feedback loops, and checklists. | 5 / 5 |
Progressive Disclosure | No bundle files exist (no references/, scripts/, or assets/), so the entire skill is one ~640-line monolith; deep-dive material (record-cut mechanics, Hister setup, embedder resolution order, limits tuning) that belongs in separate reference files is inlined. Section headers and tables do provide real structure and docs/minsync-setup.md / docs/datasource-skills.md are referenced, but only as buried inline parentheticals pointing outside the bundle — "some structure; references present but not clearly signaled; content that should be separate is inline" rather than the minimal-structure anchor below. | 3 / 5 |
Total | 16 / 20 Passed |