Content
78%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.
The body is an unusually tight, well-sequenced routing procedure with concrete tool invocations and an explicit termination condition — strong on token efficiency and actionability for an instruction-only skill. Its one real defect is the broken reference to loop-selection.md, which sends the reader to a file that does not exist in the bundle.
Suggestions
Fix the dangling reference: either ship loop-selection.md in a references/ directory or inline the loop-selection guidance (it is only needed for iterative work), since no bundle files exist at all.
Add one short worked example of an open choice and its resolution (e.g., 'Postgres vs. SQLite — resolved to SQLite because the app is single-user, Postgres lost on ops burden') to make the 'state each resolved choice with its reason and what lost' step concrete.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean with zero padding: no explanations of concepts Claude already knows, and every clause carries load (e.g., 'however likely the assumption, is open' is a load-bearing qualifier, not filler). It fully assumes Claude's competence. | 5 / 5 |
Actionability | The guidance is mostly executable for an instruction-only skill: named skills to invoke with the Skill tool (judge, draft, elicit, clarify, explain, inversion, intent, deepen), a concrete lookup list (CLAUDE.md, project rules, repo, docs, web), and a specific tool call (AskUserQuestion). It falls short of 5 because no examples illustrate what an open choice or a resolved-choice statement looks like in practice. | 4 / 5 |
Workflow Clarity | A clear sequence is present (hold changes → establish the bar → route each open choice → re-enter routing until stable → ask all at once → state resolutions) with an explicit feedback loop ('Route again until every open choice has the user's decision or confirmation and none conflict') and a verification step ('state each resolved choice with its reason and what lost'). Minor gaps — the ordering between the Route loop and 'The path' section is implicit — keep it below 5. | 4 / 5 |
Progressive Disclosure | The sections are well organized for a sub-50-line skill, but its single external reference — 'read `loop-selection.md`' — is dangling: no references/ directory or loop-selection.md file exists anywhere in the skill bundle. A broken navigation path is worse than a merely unclear signal, so this lands at 3 rather than 4. | 3 / 5 |
Total | 16 / 20 Passed |