Content
85%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-engineered overview: the body carries only selection logic, budgets, and rules, with all detail correctly pushed one level deep into 60+ verified reference files, and workflows (create, import, export) are explicitly gated by checklists and runnable verification scripts. The main deductions are deliberate cross-referenced repetition across §4/§6/§9 and two verification scripts (verify-geometry.py, verify-motion.py) that are referenced but absent from this bundle.
Suggestions
Ship scripts/verify-geometry.py and scripts/verify-motion.py in the bundle (or fold their checks into self_check.py) so the §6/§9 verification commands are runnable from the installed skill, not only a repository checkout
Replace the §4 anti-patterns row that restates the six connector rules (and the item-by-item §9 restatement) with pure rule-number back-references, e.g. 'Any breach of §6 rules 1–6'
Rewrite §0's gate paragraph as a small decision table (marker present / profile header / custom tokens / all defaults → action) to cut its dense conditional prose
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and rule-driven — tables, exact thresholds, and routing logic with no explanation of concepts Claude already knows (it never explains what a Sankey or state machine is). Minor over-explanation/replication exists: the six §6 connector rules are restated in the §4 anti-patterns table and re-listed item-by-item in the §9 checklist, typography rules appear in both §5 and §9, and §0's gate prose is convoluted (long compound sentences about marker/header/default states). Fits anchor 4 ('minor instances ... could be trimmed'), not 5 ('every token earns its place'). | 4 / 5 |
Actionability | Highly concrete for an instruction-only skill: runnable commands ('python3 scripts/self_check.py <file>', 'python3 scripts/drawio_extract.py <input>'), exact geometry values (r=8 elbows, 12px attach-point spacing, dashed 5,4, rx=6, 4px grid), and a dial table with defaults. Minor gap: 'scripts/verify-geometry.py' and 'scripts/verify-motion.py' are invoked for verification but are not shipped anywhere in this bundle (only the six other scripts exist), so two checklist commands cannot actually be run from the installed skill. Fits anchor 4 ('concrete code or commands with minor gaps'), just short of anchor 5. | 4 / 5 |
Workflow Clarity | Multi-step flows are clearly sequenced with explicit validation: §10's create-a-diagram steps (copy variant → choose pattern/type → replace → run the §9 taste gate), §11's import workflow (extract via script → set four dials → redraw → report fidelity ledger, with 'never invent ... never silently drop'), the §3 confirm-before-drawing checkpoint, and the §9 checklist acting as an explicit gate backed by runnable self-check scripts. This matches anchor 5: clear sequence, explicit validation steps, and checklist-based feedback for a complex process. | 5 / 5 |
Progressive Disclosure | Scored against the actual bundle: every one of the ~60 reference paths cited in the body (45 type-*.md files, primitives/style-guide/onboarding/output-spec/layout-budget/import/export/animation/profiles) resolves to a real file, and asset template/example patterns exist. References are one level deep and well-signaled — each link states what the target owns ('That reference owns failures, the protected default, and save behavior'; 'Full text and edge cases: ...') — and the body keeps only thresholds and routing inline while detail lives in the references. Matches anchor 5 ('clear overview with well-signaled one-level-deep references ... easy navigation'). | 5 / 5 |
Total | 18 / 20 Passed |