Content
82%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 tightly written, repo-specific review checklist that adds genuinely non-obvious convention knowledge without padding, with concrete commands and helper references throughout. It is strong on conciseness and actionability; the main gaps are the lack of an explicit validate-fix-retry feedback loop and slightly exceeding the simple-skill line for the top progressive-disclosure and workflow-clarity anchors.
Suggestions
Add an explicit review workflow sequence with a validation checkpoint (e.g. run `make fmt` then `make lint`, then `go test ./...` capturing status before any pipe) so destructive/batch-style review guidance has a clear feedback loop.
For import-grouping and similar checks, name the exact fix command (`make fmt` / `scripts/fmt-go-imports.sh`) alongside the 'flag when' condition so the guidance is copy-paste actionable.
Add a short table-of-contents or section anchor links at the top to aid navigation now that the body is ~80 lines.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Lean throughout: opens by carving out exactly what a generalist reviewer already knows ('Ordinary Go correctness, nil handling, and error wrapping are already covered without this skill') and never pads. Uses a compact 'Flag when' table and concrete examples ('honours', 'colour') with every token earning its place, assuming Claude's competence. Not below 5 — no concept over-explained. | 5 / 5 |
Actionability | Gives concrete, executable signals — exact commands (`make lint`, `make fmt`, `go run ./test/fixtures/internal/gentestdata`), exact helper calls (`tu.SkipShort(t, true)`, `tu.SkipNoNetwork(t)`), and exact envar names (`SQ_TEST_SRC__*`). Mostly executable with specific examples; not a 5 because much of the guidance is 'flag when' conditions rather than copy-paste runnable code, and a few checks (e.g. import grouping) describe the problem more than the precise fix command. | 4 / 5 |
Workflow Clarity | Not a sequenced multi-step workflow but a well-organized review checklist: a 'Highest-value checks' table with explicit 'Flag when' conditions, plus ordered guardrails ('must be run before make lint', 'Capture the status before piping'). Clear sequence with most checkpoints present; not a 5 because there is no explicit validate-then-fix feedback loop and the skill is over the 50-line simple-skill threshold, so the simple-skill 5 exception does not strictly apply. | 4 / 5 |
Progressive Disclosure | Well-organized into clearly headed sections (Highest-value checks, What lint does and does not catch, Flaky tests, Generated fixtures, Test evidence, Other conventions) with a single one-level-deep, clearly signaled reference to the canonical `AGENTS.md`. No bundle files exist, so references resolve correctly. Not a 5 because at ~80 lines it slightly exceeds the 'under 50 lines, well-organized → 5' simple-skill bar and offers no in-skill navigation/TOC. | 4 / 5 |
Total | 17 / 20 Passed |