Content
77%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.
An exceptionally dense, deeply actionable verification skill: every command, flag, exit code, and triage decision is concrete, and the workflow has genuine validation and feedback loops. Its weaknesses are that all reference-grade material (findings catalogs, deep-run statistics, coverage tables) is inlined in one large file, and the body carries time-sensitive dates, timings, and load averages that inflate tokens without aiding action.
Suggestions
Move the Findings catalogs (the 13 gate causes, 3 code-block causes, and 2 deep-run causes with their counts and example identities) into a references/ file, keeping only the cause summaries and a pointer in SKILL.md.
Drop or relocate the historical performance data ('209 s of processor time', '26 s at a load average of 42', '703 s at a load average near 130') and the fix-date ranges into a baseline/maintenance note, since they are time-sensitive and not needed to run or triage the check.
Consider moving the coverage-record table and its sync procedure to a separate file referenced from SKILL.md, so the main file reads as an overview of run, triage, and update flows.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The telegraphic style assumes competence and never explains known concepts, but it carries substantial time-sensitive data outside any deprecated section (fix dates '2026-10-24 and 2026-11-02', timings like '209 s of processor time... 26 s at a load average of 42', '703 s at a load average near 130') and a 13-cause findings catalog with per-cause counts that could be tightened or trimmed. This fits 'mostly efficient but could be tightened' rather than anchor 4, where the trimmable material would be minor. | 3 / 5 |
Actionability | Guidance is fully executable: 'pnpm check:delete-paste-ranges', '--blocks=3 --lines=3', '--lower', "--case='<identity or part of one>'", '--jobs=<n>', exit codes 0/1/2, and copy-paste git commands ('git log --oneline -1 --grep "range check over a code block and bold"'). Concrete failure identities to reuse are given, matching the copy-paste-ready anchor. | 5 / 5 |
Workflow Clarity | The batch-verification workflow is fully sequenced with explicit validation and feedback loops: gate exit codes and baseline case-count pins ('fails when either baseline's documents give a case count other than its pin'), a three-way failure classification (real bug / modeling error / intended change), '--case' for error recovery, and the loop where 'the gate fails until you run --lower'. The batch-operation cap does not apply because validation checkpoints are thorough. | 5 / 5 |
Progressive Disclosure | No bundle files exist, so all content sits inline in one ~23KB SKILL.md with clear headers but no references. Reference-grade material — the 13-cause findings catalog with counts, deep-run statistics, mutant results, and the coverage-record table — is inlined rather than split out, matching the anchor 'content that should be separate is inline'. Not anchor 4 because the split-worthy material is substantial, not a minor gap. | 3 / 5 |
Total | 16 / 20 Passed |