Content
77%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, workflow-excellent operational skill with copy-paste-ready commands and rigorous delivery gates, weakened mainly by token efficiency (three core rules each restated 4x across sections) and progressive disclosure (a 539-line monolithic body whose referenced reference/ files are absent from the bundle). Actionability and workflow clarity are near-ideal for a CLI-driven build skill.
Suggestions
State each core rule once and cross-reference it: collapse the ghost-discipline rule (currently in §Mental Model, §Ghost Discipline & Actor Lifecycle, the M-2 pitfall row, and Known Issues M-2) into a single authoritative section, and do the same for the spatial-variety thresholds and renderer caveat.
Ship the referenced bundle files (reference/decision-rules.md, reference/morph-helpers.py, reference/morph-helpers.sh, reference/styles/INDEX.md) or remove/redirect the pointers — every gate and recipe leans on helpers the reader cannot open.
Move the 52-style library lookup workflow and the full pitfall/known-issue tables into reference files, keeping SKILL.md as a pointer-plus-essentials overview.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly dense, load-bearing operational content (CLI quirks, gates, tables) with no re-teaching of concepts Claude knows, but substantial internal duplication inflates the token budget: the per-slide ghost/re-ghost rule appears in four places (Mental Model bullets, §Ghost Discipline & Actor Lifecycle, pitfall table M-2 row, Known Issues M-2), the spatial-variety thresholds (≥5cm / ≥15° / ≥30%) in four places, and the renderer caveat in four. Not 2 — sections are individually tight and non-padded; not 4 — the cross-section repetition of three core rules is a real tightening opportunity in a 539-line file. | 3 / 5 |
Actionability | Fully executable throughout: copy-paste-ready bash and Python recipes with exact props ('set "/slide[N]" --prop transition=morph', 'transition=morph-slow'), grep-verified flag claims against 'officecli help', runnable jq verification loops, and a complete Gate 2-morph leak-grep. Specific examples cover the common cases (4 recipe patterns plus adjustments). | 5 / 5 |
Workflow Clarity | Clear sequence (read pptx v2 → brief.md planning table BEFORE code → recipes → Gate 1–5a + 5b-morph-1..4 → outputs/delivery) with explicit validation checkpoints and feedback loops: every gate prints OK/REJECT, the COUNT-then-if/else enforcement pattern is taught, and 'never deliver with a known-open gate' plus re-verify-after-swap guidance close the loop. | 5 / 5 |
Progressive Disclosure | References are clearly signaled and one level deep (§References lists decision-rules.md, pptx-design.md, morph-helpers.py/.sh, styles/INDEX.md with purpose annotations), but none of the referenced files exist in the skill's bundle, and the 539-line body inlines substantial content that clearly belongs in separate reference files (the 52-style library lookup workflow, the full pitfall and known-issue tables). Not 4 — missing bundle targets plus monolithic inline bulk are more than minor gaps; not 2 — the §References section is explicit, well-organized, and not buried. | 3 / 5 |
Total | 16 / 20 Passed |