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.
An operationally excellent skill body: exact tool semantics, pointer-level paths, a decision matrix mapping user intent to operation, explicit error-recovery loops and a closing validation checklist. Its only weakness is mild redundancy — the read-before-write discipline and preview budget are each repeated across sections and could be consolidated.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with tool-specific facts Claude could not know (exact JSON pointer paths, `detail:"source"` pagination at 12,000 characters, `add_element` inserting at `afterId` or `index`) with essentially no general-knowledge padding. It edges below the top anchor because the read-fresh-before-write discipline is restated across "Read before every change", "Read fresh, then write" and the closing pass, and the preview budget appears in both "Look → edit → look, on a budget" and "Budget". | 4 / 5 |
Actionability | Guidance is fully executable: concrete tool calls with exact parameters (`read_stage path:/scenes/<order|id>` with `detail:"source"`, `str_replace` on decoded field values), exact pointer paths (`/content/canvas/elements/N/content`, `/actions/N/text`, `/content/questions/...`), and an intent-to-operation matrix covering the common edit cases. Per the rubric's instruction-skill note, absence of code blocks is not penalized when the guidance is this concrete. | 5 / 5 |
Workflow Clarity | The sequence is coherent and complete: enter with `list_scenes`, `read_stage` before every write, the smallest-operation matrix, budgeted `render_scene_preview` verification, `generate_tts` after speech rewording, and a numbered consistency-pass checklist. Explicit feedback loops for this destructive, shared-editing context are present — "A rejected patch changed nothing... Re-read and resubmit; never force" and anchor-rejection recovery via the returned diagnostic. | 5 / 5 |
Progressive Disclosure | No bundle files exist to organize; the deep reference material is correctly deferred to named sibling skills with explicit load conditions — "load `slide-craft`" for the design law and "that is `slide-dsl`" for field semantics — one level deep and clearly signaled. The body keeps only the operational core, well-sectioned with headers, a decision matrix and checklists for easy navigation. | 5 / 5 |
Total | 19 / 20 Passed |