Content
85%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
This is a well-structured hardening/quality skill with strong actionability through specific CLI commands and a clear validation ladder. The workflow is well-sequenced with explicit fail-fast checkpoints and feedback loops. The main weakness is moderate verbosity—some content is repeated across sections (e.g., redaction constraints, routing boundaries) and the Philosophy section could be tighter, though the overall token cost is reasonable for the complexity of the task.
Suggestions
Consolidate duplicated constraints (e.g., 'redact secrets' appears in both Constraints and Failure mode sections) to reduce token overhead.
Tighten the Philosophy section—the two contract references with long relative paths and 'Read when:' guidance would fit better in the Context routes section where similar references already live.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The skill is reasonably structured but contains some redundancy (e.g., the philosophy section restates principles that could be implicit, constraints like 'redact secrets' appear twice, and some sections like 'Anti-patterns' and 'Gotchas' overlap). The references to external contracts in the Philosophy section are verbose with full relative paths repeated. However, it generally avoids explaining concepts Claude already knows. | 2 / 3 |
Actionability | The skill provides specific, executable validation commands (e.g., `./bin/ask skills audit <target-skill-path> --level strict --json`), a clear ordered workflow, concrete deliverable schema fields, and explicit pass/fail/blocked outcome categories. The validation section gives exact command sequences and a defined execution order. | 3 / 3 |
Workflow Clarity | The workflow is clearly sequenced (9 numbered steps), includes explicit validation checkpoints ('fail fast: stop at the first failed gate, fix it, and rerun'), defines a specific gate ordering (strict audit → smoke eval → Plugin Eval → Tessl review), and includes feedback loops for error recovery. The failure mode section adds additional guardrails including a cap on unchanged reruns. | 3 / 3 |
Progressive Disclosure | The skill uses a well-structured 'Context routes' section with clear 'Read when:' signposts pointing to five reference files for deeper detail. The main SKILL.md serves as a map with appropriate depth, and the See Also table provides clear routing to related skills. References are one level deep and clearly signaled. | 3 / 3 |
Total | 11 / 12 Passed |