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.
A model router skill: concise, decisive, and fully executable, with unambiguous routing criteria and a disciplined handoff. The only weakness is in the surrounding bundle rather than the file itself — a large orphaned reference file and a second level of nesting.
Suggestions
Reference references/api-reference.md from the playbook (or playbooks) that needs it — it is currently 60KB of content no file points to, making it undiscoverable.
Consider surfacing the second-level reference (references/headless-rendering.md) as a clearly labeled 'deep dive' in the playbook so the nesting depth is explicit to the reader.
Add a one-line note in the handoff section that deeper references (headless rendering, API reference) exist so Claude knows the full bundle map without reading them.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~40-line body is lean with no filler: every line does routing work ('Route the current request before inspecting files, planning, invoking a specialist, or editing code'), and the disambiguation bullets ('Observable workspace state wins over wording') are substantive rules rather than padding. It assumes Claude's competence and never explains concepts Claude already knows. | 5 / 5 |
Actionability | Fully actionable for an instruction-only skill: exact file paths ('planner path: references/planner.md'), a concrete handoff sequence, and explicit prohibitions ('Do not read project files, call Bash, create a plan, or invoke a specialist between this router and the selected playbook'). The routing criteria are decidable from observable workspace state, so nothing is left abstract. | 5 / 5 |
Workflow Clarity | The routing task is simple and single-purpose, and per the simple-skills note it scores 5 when the single action is unambiguous — here the decision procedure includes explicit precedence rules ('Observable workspace state wins over wording', 'explicit user selection of a child route overrides automatic routing') and a guarded handoff ('Never load the other playbook later, including after context compaction'). No destructive or batch operations exist, so the validation cap does not apply. | 5 / 5 |
Progressive Disclosure | The entry point itself is exemplary — a router with well-signaled one-level references (planner.md, iterate.md) and 'load exactly one playbook' discipline. Scored against the actual bundle, though: references/api-reference.md (60KB) is never referenced by SKILL.md or either playbook (orphaned and undiscoverable), and the playbooks add a second reference level (references/headless-rendering.md), so navigation to part of the bundle is broken. Not 5 because content is not appropriately split across the whole bundle; not 3 because the entry file's own structure and signaling are strong and references that exist are clearly signaled. | 4 / 5 |
Total | 19 / 20 Passed |