Content
78%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 well-structured guideline skill: prescriptive decision tables, a concrete checklist, clearly separated modes, and clean one-level-deep progressive disclosure into three real reference files. The main deductions are version-conditional notes scattered through the tables and the near-total absence of inline executable code, which lives in the references instead.
Suggestions
Consolidate the Go-version-conditional guidance ('Go 1.19+', 'Go 1.21+', 'Go 1.23+', 'Go 1.25+') into a single version-gated section or the relevant reference file so the core tables stay version-stable and token-lean.
Inline one short canonical code snippet in SKILL.md (e.g., the errgroup-with-context pattern that the tables repeatedly point to) so the most common case is executable without opening a reference file.
Make the write-mode workflow an explicit ordered sequence (design → implement → `go test -race ./...` + goleak verification) with a validation checkpoint, mirroring the clarity the audit mode already has.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense, prescriptive decision tables and rules with no padded prose, but version-conditional notes are scattered throughout ('Go 1.25+: prefer `wg.Go(func(){ ... })`', 'Go 1.21+: `OnceFunc`, `OnceValue`', 'Go 1.19+ typed atomics') rather than consolidated in a versioned/deprecated section, which the rubric penalizes for time-sensitive information. | 4 / 5 |
Actionability | Concrete commands are present ('go test -race ./...', the pprof curl endpoints, 'go.uber.org/goleak') and every mistakes-table row pairs a problem with a specific fix ('Use `errgroup.SetLimit(n)`', 'Reuse `time.NewTimer` + `Reset`'), but the body itself contains almost no executable Go code — it delegates examples to the reference files — leaving minor gaps in the inline guidance. | 4 / 5 |
Workflow Clarity | Three clearly delimited modes (Write / Review / Audit), a numbered five-step parallel audit procedure, and a pre-spawn checklist ('How will it exit? ... Who owns the channels?') give a clear structure, and verification via '-race' and goleak is mentioned; however the write-mode sequence is implicit (principles and tables rather than ordered steps) and there are no explicit validate-and-retry feedback loops. | 4 / 5 |
Progressive Disclosure | The body stays an overview (principles, decision tables, checklists) and links each of the three existing bundle files from its matching section with descriptive anchor text ('For detailed channel/select code examples, see [Channels and Select Patterns](references/channels-and-select.md)'), keeping references one level deep and easy to navigate. | 5 / 5 |
Total | 17 / 20 Passed |