Content
71%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 highly actionable, well-sequenced workflow skill with excellent concrete guidance (tool schemas, prompt templates, decision trees, checklists). Its weaknesses are redundancy (preset descriptions and draft-recovery rules each appear twice), advisory-only validation around destructive index operations, and bundle references — the generate-preset sub-skill and all preset source files — that point at files not actually present.
Suggestions
Deduplicate the preset catalog: keep the style summaries in the Built-in Presets table only, and generate the Step 4.1 question option text from preset IDs plus one-line summaries instead of repeating full descriptions; likewise collapse the "Draft Context Recovery" section into a pointer to Step 0.
Fix the bundle references: either ship the presets/<platform>/<preset>/ source files and generate-preset/SKILL.md the workflow depends on, or remove/replace those paths with whatever mechanism actually supplies preset sources.
Add an executable verification step after destructive and batch operations (folder renames, delete_files, magic.project.js edits) — e.g. invoke scripts/validate_skill_contract.py or re-read magic.project.js to confirm the entry still resolves — replacing the "re-open post.json mentally" self-check.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and task-specific with no padding about concepts Claude already knows, but it could be tightened: the 20 preset style descriptions are repeated almost verbatim in both the "Built-in Presets" table and the Step 4.1 question blocks, and the "Draft Context Recovery" section re-lists Step 0's draft.json / reference-index / brand-config / articles[] extraction rules nearly word-for-word. This matches "mostly efficient but includes some unnecessary explanation or could be tightened" better than anchor 2, since the repetitions serve distinct access points (lookup table vs. copy-paste prompt). | 3 / 5 |
Actionability | Fully executable throughout: a runnable run_sdk_snippet example, complete parameter tables for both tools with types and defaults (e.g. register_in_project "Default true"), exact directory layouts, ready-to-render XML question blocks, a concrete four-layer tags JSON example, and explicit decision orders for canvas sizes and tool selection (generate_image vs image_search vs visual_understanding). Copy-paste-ready guidance covers the common cases. | 5 / 5 |
Workflow Clarity | Steps 0-6 are explicitly sequenced ("Do not skip or reorder steps"), with real checkpoints such as the "🔴 CHECKPOINT · Rednote tag confirmation", the 4.4.2 human-writing self-check list, and the failure-modes reference to consult "Before submitting". It falls short of anchor 5 because validation is advisory rather than executable — "re-open post.json mentally" — and destructive/batch operations (delete_files on posts/, folder renames, editing magic.project.js) have no verify-after-write step or feedback loop; the bundled validate_skill_contract.py script is never invoked. | 4 / 5 |
Progressive Disclosure | The References table with per-step "when to load" triggers is well signaled and the eight references/*.md files exist one level deep, but the body's core paths are broken against the actual bundle: "generate-preset/SKILL.md (inside this skill folder)" and the entire "presets/<platform>/<preset>/<preset>.css|.js" tree referenced by Step 4.1 and the source-path listing do not exist in the bundle, and scripts/validate_skill_contract.py is present but never referenced. Navigation collapses exactly where the workflow depends on it, which is more than the "minor organization gaps" of anchor 4. | 3 / 5 |
Total | 15 / 20 Passed |