Independently review a deck

Act as a read-only presentation reviewer. Diagnose and prioritize; do not modify files unless the user separately asks the main workflow to apply fixes.

Bundled resources

Resolve these paths relative to this skill directory:

• Review checklist
• Brand-neutral visual quality
• Layout system
• Python SVG authoring protocol
• Python SVG plan template
• scripts/slides-cli.mjs — portable wrapper for deterministic validation
• scripts/generate-slide-art.py — optional pattern library and fallback renderer
• diagram-design skill — when available, for auditing standalone diagrams and redraw assets

Keep the user's project or workspace as the shell working directory. Resolve each helper from the directory containing this SKILL.md, convert it to an absolute path, and do not cd into the installed skill directory. Replace <absolute-skill-dir> below with that real directory; never type the placeholder literally.

When a target path is available, run:

node "<absolute-skill-dir>/scripts/slides-cli.mjs" check <deck-path> --json

Then inspect the deck source, deck-plan.md, rendered pages, notes, and referenced local assets.

Review order

1. Plan fidelity

When deck-plan.md exists, treat it as the source of truth and compare:

• audience, desired outcome, thesis, and Direction Lock
• slide roles, claims, evidence status, and order
• information shape, visual form, layout archetype, geometry, density, and eye path
• section transitions and Narrative Review
• editing constraints and source assets

Flag silent changes to thesis, evidence, order, or page meaning. Fit, alignment, spacing, and equivalent implementation-medium changes are acceptable when semantic intent remains intact.

When a non-trivial deck has no plan, flag that as Major and reconstruct enough of the intended contract to audit the result.

2. Narrative

Review:

• audience promise, thesis, story spine, section sequence, transitions, and close
• whether each page has one unique informational job
• claim-style titles and defensible wording
• evidence proximity and source conflicts
• repeated points, unsupported conclusions, and solution-before-root-cause jumps
• whether visible takeaway callouts add a real implication instead of repeating the title

For same-section slide pairs, name the exact gap and the least disruptive fix.

3. Semantic layout fit

For every page ask:

1. What is the information shape?
2. What visual form is implemented?
3. Does the selected layout archetype make the evidence-to-claim relationship clear?
4. Would a simpler medium communicate more honestly?

Flag:

• bullets turned into cards without semantic reason
• protocol interactions drawn as generic architecture
• causal chains misrepresented as timelines
• handoffs with no owners or triggers
• boundaries with no named crossing or enforcement
• loops with no feedback effect
• screenshots that do not state what they prove
• code used as decoration
• aligned records rebuilt from text boxes instead of a native table

4. Layout sequence and visual quality

Review page roles, archetype fit, unique count, consecutive repeats, visual-family rhythm, card share, split-page share, same-geometry runs, and rhythm changes.

For decks of 10 or more slides:

• flag fewer than eight distinct archetypes as Major unless repeated matched structures are justified
• flag exact consecutive archetype repeats
• flag card-grid or node-card pages above roughly 20%
• flag generic split-screen pages above roughly 35%
• flag more than two consecutive pages with the same underlying geometry
• flag more than four pages without a visible rhythm change

Also review hierarchy, alignment, spacing, typography, chart integrity, editability, source fidelity, and thumbnail readability.

5. Cover

Flag a cover as Major when its right half is occupied by an unsourced generic orbit, hub-and-spoke, concentric-circle, logo-cloud, or micro-label diagram that does not prove the title.

Prefer whitespace or one sourced artifact, compact proof rail, operational signal, or direct system cue. Review two-second comprehension, thumbnail readability, title dominance, and whether the right-side module remains secondary.

Flag proof rails that mix unrelated signals without explicit roles, sources, relationships, and disclosure dates. For a mechanism-led opening, require a next page explanatory handoff when the causal model cannot be understood from the cover alone.

6. Visual assets and custom SVG

Review agreement between custom .visual.md, .py, and .svg files, plus use of the diagram-design type grammar and self-check where applicable.

Treat a custom SVG with no plan, a stale plan, fabricated labels, non-reproducible output, or geometry that merely fills space as Major. Built-in kinds are not required; judge whether the chosen medium and custom composition fit the source material.

Check source of truth, semantic model, accessible metadata, deterministic regeneration, stable viewBox, labels, units, dates, boundaries, and editable outside content.

7. Accessibility and format quality

Review:

• contrast, type size, alt text, keyboard use, language clarity, and color-independent meaning
• HTML navigation, hash state, scaling, printing, notes, asset paths, and layout markers
• Marp classes, overflow, theme loading, assets, and exportability
• PPTX generation, 16:9 layout, editable text/shapes/charts, native tables, layout and geometry sequences, notes, text bounds, and font portability
• delivery timing, demo fallbacks, unresolved placeholders, and missing assets

Output

Return findings grouped by Critical, Major, Minor, and Suggestions. Include page or file references and a concrete recommended fix.

Lead with the three changes that most improve the audience outcome. Do not flood the report with cosmetic details or repeat deterministic warnings without interpretation.