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 strong, unusually dense internal policy skill: nearly every line carries repo-specific evidence (scripts, guards, historical failures) with concrete commands and a dedicated validation section for CI changes. The main improvements are trimming time-sensitive dated rationale and splitting one or two detail-heavy sections into references.
Suggestions
Move the dated runner-pool figures (2026-09-29 median wait, October 2026 Enterprise Cloud migration) and the 'Real failures this replaces' quotes into an 'old patterns' / history note so the live guidance does not stale.
Consider splitting the CI-job rules table into a references/ file (like design-e2e.md) so SKILL.md reads as an overview with the per-rule evidence one level deep.
In 'Proving a CI change', state the error-recovery loop explicitly (e.g. what to do when classifier output disagrees with the expected on/off/full result) to upgrade the bullets to a validate-fix-retry sequence.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with repo-specific facts Claude could not know (script names, guard names, runner-pool numbers) and never explains generic concepts, matching the 'efficient; minor instances that could be trimmed' anchor. The dated capacity figures ('on 2026-09-29 the Fast tests gate waited a median 445 s', 'moved to Enterprise Cloud in October 2026') and the quoted historical complaints in 'Real failures this replaces' are time-sensitive rationale that will stale and could be trimmed or moved to a deprecated/old-patterns note. | 4 / 5 |
Actionability | Guidance is fully concrete and executable for an instruction-only skill: exact commands ('node --experimental-strip-types --test scripts/ci-change-scope.test.ts ...', 'pnpm mutation:changed'), exact file paths ('templates/calendar/server/lib/series-recurrence.oracle.spec.ts'), specific guard names ('guard:no-oracle-loosening', 'guard:mobile-build-paths'), and copy-ready YAML ('group: <name>-${{ github.event.pull_request.number || github.ref }}'). It is not a 5-with-gaps: the specific examples cover the common cases the skill itself names. | 5 / 5 |
Workflow Clarity | Each task area has a clear sequence and real validation checkpoints — 'Proving a CI change' specifies classifier test cases, real main-range runs with before/after evidence, local run: snippet checks, and actionlint, and the E2E section is an explicit checklist. It falls short of the 5 anchor because the validation steps are parallel bullets rather than a sequenced validate→fix→retry loop, and error-recovery paths (e.g. what to do when the classifier output is wrong) are implicit. | 4 / 5 |
Progressive Disclosure | The single bundle reference is well signaled, one level deep, and verified to exist and be self-contained ('Design's harness ... and its traps are in references/design-e2e.md'), and section headers make the body easy to navigate. Below the 5 anchor because the body carries substantial full-detail guidance inline (the CI rules table, test-worker tuning) that an ideal overview-plus-references split would move out, leaving SKILL.md as more of a full manual than an overview. | 4 / 5 |
Total | 17 / 20 Passed |