Content
82%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 unusually tight, high-signal routing skill: concrete decision tables, an executable verification command, explicit defaults, and well-handled edge cases (unconfirmed ratings, stale sessions, plugin templates). The main costs are time-sensitive statistics embedded in the main flow and a decision procedure distributed across many sections that the reader must assemble into an order.
Suggestions
Move the September–October 2026 dispatch statistics and the pointer to analyse-sessions into a short evidence/reference section (or separate file) so the main decision flow stays evergreen.
Add a brief ordered checklist of the decision checks (task demands → confirmed complexity label → DESIGN.md question → definition availability in this session → output format) so the reader doesn't have to infer precedence across sections.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and purposeful — "Defaulting up is cheap and recoverable; defaulting down produces a subagent that quietly does a worse job and reports success" — with no explanations of concepts Claude already knows. It misses anchor 5 because time-sensitive statistics ("Over September and October 2026, about half of all dispatches...") sit in the main flow rather than an old-patterns section, and a few table cells (the reviewer row) carry more context than the decision needs. | 4 / 5 |
Actionability | For an instruction-only skill the guidance is fully actionable: a decision table mapping each agent definition to concrete task types, an executable check (`grep -n '"<package>"' <workspace>/package.json`), an explicit default ("default to `implementer`"), a defined output format ("One name... and one sentence saying why"), and precise exception rules (e.g. the effort:xhigh carve-out on the full check tier). Concrete specific guidance covers the common cases, matching anchor 5 under the instruction-only scoring note. | 5 / 5 |
Workflow Clarity | The single action (pick a definition, justify it in one sentence) is unambiguous and carries built-in checkpoints — "Read the file of the one you choose rather than assuming either" and "Check the semver delta before you choose, do not infer it". It stops short of anchor 5 because the decision procedure is spread across many sections (rubric table, complexity table, design-agent rules, the general-purpose ban, the plugin exception, the stale-session fallback) with no explicit ordering, so the reader must synthesize the sequence themselves. | 4 / 5 |
Progressive Disclosure | Sections are well-organized with clear headers and tables, and there are no nested or buried references — the cross-references to other skills (orchestrate, prep-pr, analyse-sessions) are clearly signaled. Anchor 4 fits rather than 5 because the skill exceeds the under-50-line simple-skill case and inlines contextual material (the dispatch-site map, historical dispatch statistics) that could live in a separate reference file. | 4 / 5 |
Total | 17 / 20 Passed |