Content
55%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
This skill is exceptionally thorough and actionable with excellent workflow clarity, validation checkpoints, and concrete executable commands throughout all phases. However, it is severely over-long for a single SKILL.md file — the monolithic structure with repeated token resolution blocks, massive inline sub-agent prompts, and no progressive disclosure makes it extremely token-inefficient. The content would benefit enormously from being split into a main overview with referenced sub-files for sub-agent templates, claim tracking logic, and phase-specific details.
Suggestions
Extract the two large sub-agent prompt templates (fix agent and review handler) into separate referenced files (e.g., TEMPLATES/fix-agent.md and TEMPLATES/review-handler.md) to dramatically reduce the main file size.
Consolidate the token resolution logic (repeated 4+ times across main skill and sub-agent prompts) into a single referenced script or shared section.
Remove explanations of basic concepts Claude already knows (e.g., HTTPS vs SSH URL format parsing, how to extract owner/repo from a git remote URL) and replace with concise one-liners.
Split Phase 4 (pre-flight checks with 7 sub-steps) and Phase 6 (review handler with 6 sub-steps) into separate referenced files, keeping only summaries in the main SKILL.md.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Extremely verbose at ~600+ lines. Massive amounts of inline detail that could be split into referenced files. The sub-agent prompt templates are enormous and repeated (token resolution appears 4+ times). Much content explains things Claude already knows (how to parse JSON, how git remotes work, HTTPS vs SSH URL formats). The claim-based tracking, cursor file logic, and embedded review parsing are all inlined when they could be separate references. | 1 / 3 |
Actionability | Highly actionable with concrete, executable curl commands, git commands, and specific API endpoints. Every step includes copy-paste ready code with proper headers, URL patterns, and error handling. The sub-agent prompts are fully specified with real commands rather than pseudocode. | 3 / 3 |
Workflow Clarity | Excellent multi-phase workflow with clear sequencing (6 phases), explicit validation checkpoints (token verification, remote access checks, existing PR detection, claim-based dedup, dirty tree warnings), and error recovery paths (retry logic in tests, timeout handling, watch mode loops). Feedback loops are well-defined throughout. | 3 / 3 |
Progressive Disclosure | Monolithic wall of text with no bundle files or external references. The entire skill is one massive document. Sub-agent prompt templates (which are substantial) are inlined rather than referenced. The review handler logic, claim tracking, and cursor management could all be separate files. No bundle files are provided to support this content. | 1 / 3 |
Total | 8 / 12 Passed |