Content
87%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
An efficient, well-structured instruction skill with concrete deletables and a clear sweep sequence. The only gap is the absence of an explicit error-recovery feedback loop after the verification step, which caps workflow clarity at 2 for this destructive operation.
Suggestions
Add an explicit feedback loop after Sweep step 5: state what to do if verification fails (fix the breakage, re-run verification, only stop when it passes).
Name a concrete verification command or check in step 5 (e.g. build, typecheck, or test suite) instead of the abstract 'narrowest honest verification'.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Lean throughout — every section (Rules, Sweep, Smells, Bias) earns its place, assumes Claude's competence, and explains no background concepts Claude already knows. | 3 / 3 |
Actionability | Instruction-only but concrete: names the tool ('Trace references with `rg`'), enumerates specific deletables (surface, callers, types, state, tests, docs, comments), and gives a numbered sweep procedure — actionable per the instruction-only scoring note. | 3 / 3 |
Workflow Clarity | The Sweep section is a clear 5-step sequence with a verification step (step 5, 'Run the narrowest honest verification'), but it lacks an explicit validate→fix→retry feedback loop, which the guidelines require for destructive operations. | 2 / 3 |
Progressive Disclosure | A simple, single-purpose skill under 50 lines with no external references; its well-organized sections (Rules, Sweep, Smells, Bias) justify a 3 under the simple-skills scoring note. | 3 / 3 |
Total | 11 / 12 Passed |