Content
75%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 unusually concrete audit skill: each smell category pairs an executable detection command with a validated real example, a triage rule, and false-positive guidance, and the rejected-categories section prevents wasted hunts. The main weaknesses are minor: BRE-fragile grep patterns, a missing post-fix verification step, and some logger detail that could be split into a reference file.
Suggestions
Fix the grep patterns so they run as written: add -E (or -P) to the category 1, 2, and 7 greps that rely on \s and parenthesized alternation, which otherwise match literally and silently return nothing.
Add a final verification step to the Workflow (e.g., 're-run the greps on the changed files and confirm zero untriaged hits before opening the PR').
Move the logger injection patterns and the two consoleSink/new Error evasion greps into a references/ file (e.g., references/logging.md) to slim category 3, and deduplicate the single-method Pick grep shared by categories 5 and 6.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and assumes competence (no explanation of what TypeScript or grep are), and every category carries a pattern, triage rule, and real example. A few sections could be trimmed — the apps/whispering narrative in category 3 is long, and the single-method `Pick` grep is repeated verbatim in categories 5 and 6 — so it is not the fully lean fit of 5. | 4 / 5 |
Actionability | Every category ships a concrete grep/rg command plus a real file-path example and a step-by-step triage recipe. Not 5 because several grep patterns use BRE metacharacters (\s, bare parens, backreference-style alternation) that fail without -E/-P as written, so they are not reliably copy-paste ready. | 4 / 5 |
Workflow Clarity | The Workflow section gives a clear four-step sequence (run greps → triage each hit as justified/refactor/false-positive → group by category → single PR with audit log), and the triage step acts as a checkpoint. Not 5 because there is no explicit verification loop, e.g., re-running the greps after fixes to confirm the hits are gone. | 4 / 5 |
Progressive Disclosure | No bundle files exist, and the single file is well organized with numbered categories, a rejected-patterns section, and clearly signaled cross-references to other skills (define-errors, logging, refactoring) and one spec file. Not 5 because some detail that belongs in reference files is inlined — the logger injection patterns and consoleSink evasion prose in category 3 could live in a references/ file. | 4 / 5 |
Total | 16 / 20 Passed |