Content
70%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 an exceptionally actionable, well-sequenced orchestration playbook with validation and error recovery built into every phase, but it pays for this with heavy redundancy and a monolithic single-file layout: token-resolution, spawn-config, and cleanup notes repeat four-plus times and the sub-agent prompt templates are inlined rather than referenced.
Suggestions
Move the two sub-agent prompt templates into references/sub-agent-prompt.md and references/review-agent-prompt.md, leaving a variable list and pointer in SKILL.md.
Factor token resolution and the spawn-configuration block into a single stated-once section referenced from each phase, removing the four duplicated copies.
State the gh-CLI prohibition and the worktree-lifecycle note once in a Constraints section instead of repeating them in every prompt and phase.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~870-line body repeats the same material many times: token resolution appears in Phase 2, twice inside the sub-agent prompt (jq and node variants), and again in the review prompt; the spawn-configuration block and the "no cleanup input is needed" note each repeat four times; the "Do NOT use the gh CLI" warning repeats 5+ times. These are several unnecessary padded sections, matching the anchor for noticeable verbosity; not a 3 because the duplication is extensive rather than incidental tightening. | 2 / 5 |
Actionability | Guidance is fully executable: exact curl invocations with headers, concrete git commands (e.g. git remote add fork https://x-access-token:$GH_TOKEN@github.com/{PUSH_REPO}.git), jq filters for claims expiry, and complete verbatim sub-agent prompt templates. Copy-paste ready commands cover the common cases; no pseudocode. | 5 / 5 |
Workflow Clarity | Six explicitly numbered phases with sequencing rules ("If --reviews-only is set: Skip directly to Phase 6"), extensive validation checkpoints (dirty-tree check, token validity HTTP check, remote access, existing-PR/branch checks, claim expiry), error handling per API call (401/403 paths), and feedback loops (test-failure retry, confidence gate before implementing). This matches the top anchor including the batch-operation validation requirement; not a 4 because validation coverage is essentially complete. | 5 / 5 |
Progressive Disclosure | No bundle files exist and everything is inline in one ~870-line SKILL.md: the two ~120-line sub-agent prompt templates and the claims/cursor procedures clearly belong in separate reference files. Section structure (phases, headers, tables) is present and navigable, matching the anchor 'some structure but content that should be separate is inline'; not a 2 because the structure is not minimal, and not a 4 because the monolithic inlined prompts are a significant organization gap. | 3 / 5 |
Total | 15 / 20 Passed |