Content
92%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 a disciplined, dense operations manual: it discloses only the tool contracts Claude could not know, sequences edits with explicit read-verify loops, and pushes per-type field detail to four real one-level-deep reference chapters. The sole minor gap is the absence of one complete example tool invocation or op payload.
Suggestions
Add one complete worked example of a patch_stage call (target, intent, and a single op object) so the op-table fields can be assembled without inference.
Include one short example grep_stage call showing the truncated:true / cursor continuation loop in use, since cursor semantics are stated but never demonstrated.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and assumes competence: it explains only OpenMAIC-specific contracts (the document model tree, op semantics, pointer escaping as ~1/~0, pagination via nextOffset, NFKC normalization behavior, the 2 KiB media placeholder) — none of which Claude would already know. Tables carry the bulk of the content compactly ('Tool vocabulary', 'Addressing', op table, routing table) and there is no padded prose or restatement of generic concepts, matching the 'every token earns its place' anchor. | 5 / 5 |
Actionability | Guidance is highly concrete: exact path syntax ('path:/scenes/<order|sceneId>', '/scenes/3', '/content/questions/1/options/0/label'), a per-op field table, and literal workflow commands ('Read the target scene with detail:"source"'... 'Pass the returned nextOffset back as offset'). It stops short of a complete worked example of a full tool call or op payload, leaving a minor gap — 'mostly executable with minor gaps' (4) rather than fully copy-paste-ready (5). | 4 / 5 |
Workflow Clarity | The 'Read before write' section gives an explicit 6-step sequence with validation checkpoints ('Read the same source path again and verify the stored value', 'use detail:"text" or grep_stage when the check is "no old copy remains"'), the patch section documents atomicity with a failure loop ('If op 2 fails, op 1 is not persisted'; a rejected batch 'changed nothing'), and rejected ops route to the matching reference chapter — a clear sequence with explicit validation, feedback loops, and a hard-rules checklist, matching the 5 anchor. | 5 / 5 |
Progressive Disclosure | SKILL.md is an overview that routes by need via a one-level-deep table ('Quiz questions, options, answers, grading fields → references/quiz.md', etc.), and all four referenced files exist in the bundle (quiz.md, widget.md, actions.md, pbl.md, each a substantial single-chapter field reference). The slide-dsl material is delegated to the installed skill rather than duplicated. Clear overview, well-signaled one-level-deep references, easy navigation — the 5 anchor. | 5 / 5 |
Total | 19 / 20 Passed |