Content
81%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 well-structured operational skill: the 7-phase pipeline is clearly sequenced with genuine validation checkpoints and feedback loops, tool usage and error recovery are concrete, and detail is appropriately deferred to external files. The remaining gaps are modest — mild duplication of trigger/capability statements across intro and When-to-Use sections, a missing worked recipe example, and inconsistent external reference paths.
Suggestions
Trim the duplication between the intro paragraph, the 'Contract' section, and 'When to Use' — trigger phrases appear in the frontmatter, the intro, and When to Use, so one restatement suffices.
Include a short example recipe YAML (5–10 lines with source queries, classification rules, and extraction schema fields) so Phase 1's 'copy a built-in recipe and edit' has a concrete anchor without leaving the skill.
Unify the external reference paths to a single consistent form ('../conventions/agent-operator-protocol.md' vs 'skills/conventions/quality.md' currently mix relative and pseudo-absolute styles).
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and operational with zero filler and no explanations of concepts Claude already knows, but the intro paragraph, 'Contract', and 'When to Use' sections each restate capabilities/trigger phrases already present in the frontmatter — minor duplication that could be trimmed, placing it at the 4 anchor rather than 'every token earns its place'. | 4 / 5 |
Actionability | Guidance is mostly executable within its tool environment: concrete tool names (put_raw_data, put_page, file_upload), paths (~/.gbrain/recipes/{name}.yaml), a specific recovery command (gbrain write-request <request_id>), and a worked output-table example; minor gaps remain — no example recipe YAML and no concrete search-query samples — so it falls short of fully copy-paste-ready. | 4 / 5 |
Workflow Clarity | Seven clearly sequenced phases with explicit validation for batch/destructive operations (the extraction integrity rule mandates re-reading from saved files and never trusting LLM working memory; dedup runs before any tracker write), feedback loops for error recovery (revision_conflict → re-read and merge; rate_limited → back off and log the gap; LLM-fallback logging to improve regexes), and a numbered checklist — matching the top anchor. | 5 / 5 |
Progressive Disclosure | A well-organized overview with details appropriately split out — recipes live as separate YAML files, error-protocol and citation rules are deferred to clearly signaled, one-level-deep docs — but the two conventions references use inconsistent path forms ('../conventions/agent-operator-protocol.md' vs 'skills/conventions/quality.md') and cannot be verified against the bundle (no references/ directory ships with it), a minor organization gap versus the 5 anchor. | 4 / 5 |
Total | 17 / 20 Passed |