Content
62%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 well-sequenced, highly actionable review-loop prompt with strong validation and termination logic. Its weaknesses are redundancy (duplicated criteria list, speculative phase predictions) and a monolithic structure that inlines a 400-line template instead of splitting it into a reference file.
Suggestions
Delete the 'REVIEW CRITERIA REFERENCE (Quick Check)' section and the speculative 'Expected Behavior' phase predictions - both duplicate content already stated in Step 2 - and fold the one useful line from 'Customization' into the main workflow.
Move the ~400-line review-loop prompt into references/review-prompt.md and have SKILL.md show only the ralph-loop invocation plus a pointer, so users can cat or pipe the prompt file directly instead of copy-pasting it out of a markdown code block.
Replace '[paste prompt below]' in Quick Start with the concrete file-based invocation (e.g. /ralph-wiggum:ralph-loop "$(cat references/review-prompt.md)") so the quick-start command is executable as written.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The 'REVIEW CRITERIA REFERENCE (Quick Check)' section re-lists the same 16 criteria already enumerated in Step 2, and 'Expected Behavior', 'Why This Works', and 'Customization' add padded or speculative content with no operational value. | 2 / 5 |
Actionability | State-file JSON, init/monitor bash commands, a syntax-check loop, exact commit-message and findings templates are all concrete and near copy-paste ready. Minor gaps remain: the Quick Start requires manually assembling the '[paste prompt below]' command, and several templates are placeholder-laden ('[N]', '[count]'). | 4 / 5 |
Workflow Clarity | Steps 1-6 are explicitly sequenced with validation checkpoints (Step 4 post-improvement verification, Step 6 termination check via jq), feedback loops (review -> fix -> verify -> re-review), an escape hatch after 50 stalled iterations, and a completion checklist gating the promise tag. | 5 / 5 |
Progressive Disclosure | No bundle files exist; the ~400-line embedded prompt is inlined monolithically in SKILL.md when it belongs in a references/ file users could read or copy directly. Internal section headers do keep navigation reasonable, placing this above the 'minimal structure' anchor. | 3 / 5 |
Total | 14 / 20 Passed |