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 actionable, well-sequenced workflow skill: concrete commands throughout, explicit validation and error-recovery loops, and clear mode-dependent paths. Its main weakness is token efficiency — the required block over-repeats its core mandate and the full-clean-build guidance is stated twice — which also creates a minor progressive-disclosure organization gap.
Suggestions
State the ask-first mandate once in the <required> block and cut the redundant restatements ("Do not pick a mode for them. Do not default to a mode. Auto/autonomous mode does NOT authorize you to skip this question...") — the conservative-exceptions text already carries the enforcement weight.
Consolidate the duplicated full-clean-build guidance (Step 3 and the "Determining if the public API surface changed" section) into a single location, or move it into tree-api-checks.md with a one-line pointer from both spots.
Trim the per-mode task-listing detail in the required block to the mode→steps mapping only; the conditional tree-package cascade task is repeated in Step 6 and can live there alone.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is command-first and avoids explaining known concepts, but the <required> block restates the ask-first mandate repeatedly ("You MUST ask the user the Step 1 mode-choice question before doing anything else. Do not pick a mode for them. Do not default to a mode. Auto/autonomous mode does NOT authorize you to skip this question...") and the full-clean-build procedure ("pnpm clean && pnpm build", per-package cleans unreliable for the tree package) appears in full both in Step 3 and again in the "Determining if the public API surface changed" section. This matches anchor 3 ("mostly efficient but includes some unnecessary explanation or could be tightened") rather than 4 because the repetition and duplication go beyond minor trimming. | 3 / 5 |
Actionability | Every step carries copy-paste-ready commands with documented defaults ("bash .github/skills/ci-readiness-check/ci-readiness-check.sh [base-branch]", "cd $PKG && pnpm exec fluid-build . -t build:api-reports", "git diff --name-only HEAD -- | grep api-report"), plus concrete error-path handling ("If API Extractor fails with ae-missing-release-tag, add a TSDoc release tag... rebuild, then retry"). Matches anchor 5 ("fully executable; copy-paste ready; specific examples cover the common cases") — not 4, since error paths, mode variants, and defaults leave no actionable gaps. | 5 / 5 |
Workflow Clarity | Steps 1-9 are clearly sequenced with explicit mode gating ("Check mode stops here — skip steps 4–8 entirely"), feedback loops for error recovery ("If API Extractor fails... rebuild, then retry"; unexpected API report diffs → full clean build → "re-run the CI readiness check"), and checklists (the 5-item final status report, per-mode task lists). Validation follows the batch auto-fix operations (verify checks pass, check reports afterward, review git status), so the destructive/batch cap does not apply — a clear anchor-5 match, not 4, because no validation checkpoints are missing. | 5 / 5 |
Progressive Disclosure | The tree-package deep-dive is properly split out with a clear conditional trigger ("read .github/skills/ci-readiness-check/tree-api-checks.md before proceeding with 6a") and steps use clear numbered headers, but the duplicated inline clean-build troubleshooting could be consolidated into that reference file and the required block's task-listing detail is dense. Matches anchor 4 ("good structure; most content is appropriately placed; references mostly clear; minor organization gaps") — not 5 because of the inline duplication and the absence of a reference bundle beyond the two path-referenced files. | 4 / 5 |
Total | 17 / 20 Passed |