Content
81%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 highly operational, well-guarded skill body: every workflow step is executable, cost and safety checkpoints are explicit, and failures have concrete recovery paths. The chief weakness is redundancy — calibration defaults and agent-choice rules are stated repeatedly, and the schema-style meta sections restate each other — which inflates token cost without adding guidance.
Suggestions
State the calibration default ("--limit 50 --concurrency 5, defer to user-named values") once in the canonical workflow and reference it from Guardrails 1 and 13 instead of repeating it four times.
Consolidate the overlapping meta sections (Intent signature / When to use / Control-flow features / Preconditions / Effects) into a single compact routing-and-constraints block to cut duplicated tokens.
Rename the "Scheduling" top-level header to something accurate like "Scope and Routing" so the section tree matches its content.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and mostly token-earning (exact commands, cost bands, thresholds), but the calibration advice "--limit 50 --concurrency 5" is repeated in four places (Transitions, workflow step 2, Guardrails 1 and 13), the "ask agent choice before the first paid call" rule appears in both Entry and Guardrail 13, and the meta sections (Intent signature, Control-flow features, Resource scope, Preconditions, Effects) partially duplicate one another — matching the mostly-efficient-but-could-be-tightened anchor. | 3 / 5 |
Actionability | Every phase has copy-paste-ready bash (init, scan, process with concrete flags, triage, revalidate, export, CI command), plus specific cost bands ("100 files ≈ $25-60, 500 ≈ $130-300, 2,000 ≈ $500-1,200"), matcher hit-rate targets, and exact file paths, fully matching the executable-and-covers-common-cases anchor. | 5 / 5 |
Workflow Clarity | The canonical workflow is explicitly sequenced (bootstrap → calibrate → full pass → PR mode → matchers → resume) with genuine validation checkpoints and feedback loops: cost extrapolation plus user go-ahead before unbounded spend, quota stops with re-run-same-command resume, and an FP-rate loop (revalidate → tighten INFO.md → bias matchers precise), matching the explicit-validation anchor. | 5 / 5 |
Progressive Disclosure | The References section clearly maps each intent to one-level-deep resource files (setup, scanning, pr-review, matchers, triage, config) with labeled purposes, satisfying good structure; it is not a 5 because the ~210-line body inlines substantial detail (guardrails, failure tables, matcher workflow) that could live in those resources, and the top-level "Scheduling" header mislabels its goal/intent content. | 4 / 5 |
Total | 17 / 20 Passed |