Content
65%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 tight, well-structured instruction skill whose guidance is directionally concrete but under-specified for execution: most scan sources have no commands and the auto-fix workflow lacks validation. The untrusted-input handling is a strength appropriate to the skill's inputs.
Suggestions
Add concrete commands for each scan source, e.g. `git log --merges -20`, `grep -rnE 'TODO|FIXME|HACK' <changed files>`, and the project's actual lint command.
Add a validation checkpoint after auto-fixes (re-run lint/tests and show the diff before counting a fix attempt) to support the batch fix loop.
Specify the output format for post-merge-state.md (e.g., columns for priority, item path, classification, status) so the 'prioritized cleanup list' is unambiguous.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean (~40 lines) with no explanations of concepts Claude already knows ('Scan Sources', 'Classification', 'Rules' are all tight); the injected untrusted-input block is functional rather than padded, though minor content like the 'Background:' link could be trimmed. | 4 / 5 |
Actionability | One executable command is given ('git diff HEAD~10') plus concrete classification buckets and limits ('Max 2 fix attempts'), but most scan sources lack commands (how to list recent merges, grep TODO/FIXME/HACK, run the linter) and the 'prioritized cleanup list' output format is unspecified — missing key execution details. | 3 / 5 |
Workflow Clarity | Sections imply a scan -> classify -> output sequence, but it is never made explicit as steps, and the workflow involves batch auto-fixing with no validation/verification checkpoint (e.g., re-run lint or tests after a fix), which caps this dimension at 3 per the batch-operation rule. | 3 / 5 |
Progressive Disclosure | The skill is under 50 lines with no need for external reference files, and content is well-organized into clearly headed sections (Scan Sources, Classification, Output, Rules); 'post-merge-state.md' is an output state file, not a bundled reference, so there is nothing to disclose progressively. | 5 / 5 |
Total | 15 / 20 Passed |