Content
67%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, well-structured skill body with a clear phase sequence and genuinely strong error-handling guidance. The main weakness is actionability: the transform examples describe inputs and outputs rather than showing an executable transform function or complete commands.
Suggestions
Show one complete, executable example transform (e.g., the SMS case) with actual payload-to-brain-page mapping code or a concrete gbrain put invocation with arguments, instead of abstract Input/Output pairs.
Add an explicit validation checkpoint in Phase 3 (e.g., verify the transformed page is sanitized and well-formed before the brain write) to close the workflow's only validation gap.
De-duplicate sanitization and dead-letter guidance between the Contract and Anti-Patterns/Phase 4 sections to reclaim tokens.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and operational with no concept explanations Claude already knows, but there is minor redundancy: sanitization appears in both the Contract ('no raw HTML/script passes to brain pages') and twice in Anti-Patterns, and dead-letter handling is stated in both the Contract and Phase 4. | 4 / 5 |
Actionability | Concrete specifics exist ('gbrain put', 'gbrain sync', '_dead-letter/{timestamp}.md'), but the example 'transforms' are abstract Input/Output pairs rather than executable transform functions, and no complete gbrain command with arguments is shown, leaving key execution details missing. | 3 / 5 |
Workflow Clarity | Phases 1-4 are clearly sequenced with an explicit error-handling phase (log to dead-letter, retry once, 'Don't lose events') and a 'When it fails' section with concrete recovery guidance (poll write_pending, reject malformed payloads). It falls short of 5 because there is no validation checkpoint on the transform output before the brain write. | 4 / 5 |
Progressive Disclosure | Sections are well-organized, no content is inlined that clearly belongs in a separate file, and there is exactly one clearly signaled one-level reference. It does not reach 5 because that sole reference ('../conventions/agent-operator-protocol.md') does not resolve within the skill bundle, and the body slightly exceeds the compact-skill threshold. | 4 / 5 |
Total | 15 / 20 Passed |