Content
67%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 well-structured conventions skill with concrete, actionable rules and correctly offloaded deep reference material. Its main cost is token weight: a persuasive multi-paragraph rationale and a "The short version" section duplicate content that could be stated once, tightly.
Suggestions
Cut the rhetorical failure narrative in "Hooks live at the edge" (the "no failing test… no bisect helps" passage) down to one or two sentences — the two acceptance criteria and the Fine/Never pair already carry the rule.
Delete "The short version" section, or replace the body's long rationale with it; having both roughly doubles the core guidance for no new information.
Add one short example of the single allowed hook shape (a useSyncExternalStore wire to the engine) so the "entire allowed shape" is copy-paste concrete rather than described.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The core rules are substantive and non-obvious team policy ("The engine package must not import React", the apply/skip list, the kebab-case value examples), but there is real padding: the rhetorical failure narrative ("no failing test, no benchmark regression, no noisy error to alert you. By the time someone notices, the regression is weeks deep and no bisect helps, because nothing was ever measurable") and a redundant "The short version" section restating the body. This matches anchor 3 ("mostly efficient but includes some unnecessary explanation or could be tightened"); it is not 2 because the material is not explaining concepts Claude already knows, and not 4 because the duplication and rhetoric exceed minor trimming. | 3 / 5 |
Actionability | Concrete, executable guidance throughout: "A hook is acceptable in exactly two situations" with testable criteria, "reduce the hook to a three-line call site", "The engine package must not import React", a Fine/Never example pair, and worked testid values like "sidebar-right-inspect-node-properties". Mostly executable with minor gaps (no example of an acceptable use-* adapter), matching anchor 4; not 5 because no copy-paste example of the one allowed hook shape is shown, and not 3 because nothing is pseudocode or missing key details. | 4 / 5 |
Workflow Clarity | The decision procedure is clear and ordered: check the two acceptable hook situations, otherwise "extract the logic out, test and benchmark it there, and reduce the hook to a three-line call site", plus structural enforcement (engine must not import React). Matches anchor 4; not 5 because the acceptance criteria are subjective ("short enough that no test could catch a bug the code couldn't already reveal at sight") with no concrete verification step for checking compliance, and not 3 because the sequence and decision points are explicit, not implicit. | 4 / 5 |
Progressive Disclosure | Well-organized sections with the bulk of the testid detail correctly offloaded and clearly signaled: "The full reference — with the apply/skip table and worked examples — lives in docs/contributing/react.md", plus a one-level "See also code-ts" link. No bundle files exist, so this is scored on the body's structure alone; matches anchor 4. Not 5 because the body carries a long inline rationale and a duplicate summary section that could be trimmed, leaving the SKILL.md less of a lean overview; not 3 because structure is good and references are clearly signaled, not buried. | 4 / 5 |
Total | 15 / 20 Passed |