Content
96%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.
An exemplary workflow skill: terse, fully actionable, and sequenced with genuine validation checkpoints, feedback loops, and stop conditions. The only weakness is minor inconsistency in how cross-skill references are signaled and verified.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Phrases like 'Stage only files for this ticket; exclude unrelated local changes', 'Pick the smallest check that proves the change', and the one-line command table show a lean, checklist-driven body that assumes Claude already knows git, CI, and PR mechanics. No concept explanations or padding anywhere; every line is an instruction, guardrail, or trigger condition. | 5 / 5 |
Actionability | Concrete, copy-paste-ready guidance throughout: exact commands ('git log next..HEAD --oneline', 'pnpm --filter @novu/api-service build', 'gh pr create'), exact commit format ('type(scope): concise why fixes NV-XXX'), a change-type→command table, and an explicit CI triage decision tree ('If every failure is related, deterministic, and high confidence... fix, run local validation, commit, push' vs 'If flake... do not guess-fix'). Common cases (API changes, shared-lib changes, e2e changes) are each covered with a specific action. | 5 / 5 |
Workflow Clarity | Nine numbered steps sequenced in dependency order with a copy-track checklist, explicit validation checkpoints ('Report failures; fix before push when deterministic', merge-ready checklist before 'done'), and real feedback loops (CI fail → fix → local validation → commit → push; 'If P0 cross-tenant risk is found: fix + add e2e coverage before opening/updating PR'). Stop conditions and explicit guards on destructive actions ('Commit only when the user asks', 'Never modify CI config/workflows') close the loop on risky operations. | 5 / 5 |
Progressive Disclosure | Well-organized single-file skill with clearly signaled one-level-deep references (a markdown link to run-api-e2e-tests in the validation table, enterprise-submodule in step 6). However, sibling-skill references are inconsistently signaled — 'thermo-nuclear-code-quality-review', 'deslop', 'get-pr-comments', and 'babysit' appear only as backticked names rather than links, and none of the referenced sibling paths exist alongside this SKILL.md in the bundle, so navigation to them cannot be verified. | 4 / 5 |
Total | 19 / 20 Passed |