Content
92%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 instruction-only skill body: a five-stage workflow with explicit 'Ready when' validation gates, a failure-feedback loop, a concrete depth-selection table, and a checklist-style handoff format, all delivered in lean prose with clean one-level-deep reference routing. The only gap is that concrete worked instances and environment-specific commands are deferred to references rather than shown inline.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean, imperative, and free of padding — e.g., 'Execute one bounded unit per owner at a time, check its result, and record the observed evidence before dependent work' — with no explanation of concepts Claude already knows. Every section (role, workflow, output, references) earns its place, matching the 'lean and efficient; assumes Claude's competence' anchor rather than the 4 anchor's 'minor instances of over-explanation'. | 5 / 5 |
Actionability | Concrete, usable structures throughout: the four contract fields (Outcome, Scope and constraints, Verification, Completion condition), a Compact/Structured/Design-intensive depth-selection table with explicit 'Select when' criteria, per-stage 'Ready when' gates, and a five-field output checklist. It falls short of 5 because several directives deliberately defer specifics to the environment ('Discover concrete commands and conventions from the environment') and no worked instance is inlined (examples live in worked-examples.md); it is well above 3's pseudocode-level vagueness. | 4 / 5 |
Workflow Clarity | Five numbered stages each end in an explicit validation checkpoint ('**Ready when:**...'), stage 4 contains a full error-recovery feedback loop ('When a check fails, use its evidence to diagnose and correct the unit; when attempts repeat the same failure, revisit the premise'), and stage 5 covers interruption/resume reconciliation. The output section is itself a checklist, matching the 5 anchor (explicit validation steps, feedback loops, checklist); the destructive/batch cap does not apply since this skill coordinates rather than performs risky operations. | 5 / 5 |
Progressive Disclosure | All four bundle references (task-contracts.md, decomposition.md, verification.md, worked-examples.md) exist and are exactly one level deep. Each is signalled inline at its decision point ('Read [task contracts](references/task-contracts.md) when creating a structured or design-intensive plan...') and again in a References section with a one-line 'read for' purpose, plus the guard note 'Select the current decision; these are not additional mandatory phases.' This matches the clear-overview, well-signaled one-level-deep 5 anchor; the ../sibling-skill links are cross-family navigation, not nested bundle references. | 5 / 5 |
Total | 19 / 20 Passed |