
Slides Critique
- 151 installs
- 9 repo stars
- Updated March 15, 2026
- mpuig/agent-slides
slides-critique is an agent-slides skill that reviews PowerPoint decks for storytelling quality, action titles, narrative flow, and visual hierarchy for developers and technical leads who need persuasive presentation con
About
slides-critique is a presentation storytelling skill in the mpuig/agent-slides toolkit for reviewing output.pptx decks against Pyramid Principle, SCQA, and action-title criteria. The skill requires Python 3.12+ and uv, reads deck content via uvx agent-slides slides inspect, cross-references slides.json plans and design-profile.json palettes, and scores seven dimensions: action titles, narrative flow, isomorphism, visual hierarchy, content density, layout variety, and parallel structure. Developers reach for slides-critique after slides-build when a deck is technically valid but unconvincing, or when slide titles read as topic labels instead of claims. The workflow produces a critique report with positives, a scorecard, provocative questions, auto-applied text fixes via slides edit, and structural fixes via critique-fixes.json, then re-runs slides qa.
- Scores 7 dimensions: action titles, narrative flow, isomorphism, hierarchy, density, layout variety, parallel structure
- Uses uvx agent-slides slides inspect on output.pptx with slides.json plan cross-reference
- Auto-fixes action titles via slides edit and structural ops via critique-fixes.json
- Loads storytelling.md and common-mistakes.md reference guides before analysis
- Requires Python 3.12+ and uv; distinct from technical slides-audit linting
Slides Critique by the numbers
- 151 all-time installs (skills.sh)
- +8 installs in the week ending Aug 4, 2026 (Skillselion tracking)
- Ranked #1,016 of 1,880 Design & UI/UX skills by installs in the Skillselion catalog
- Data as of Aug 4, 2026 (Skillselion catalog sync)
npx skills add https://github.com/mpuig/agent-slides --skill slides-critiqueAdd your badge
Show developers this skill is listed on Skillselion. Paste this into your README.
| Installs | 151 |
|---|---|
| repo stars | ★ 9 |
| Last updated | March 15, 2026 |
| Repository | mpuig/agent-slides ↗ |
How do you critique presentation storytelling in PowerPoint?
Review output.pptx with slides-critique to score action titles, narrative flow, and isomorphism against slides.json before a stakeholder presentation.
Who is it for?
Developers or tech leads refining a built agent-slides deck when narrative structure and action titles need partner-level feedback.
Skip if: Teams needing font overlap, bounds, or formatting lint—that belongs to the separate slides-audit skill.
When should I use this skill?
A user says review my deck, critique the presentation, check narrative flow, or improve slide titles on an agent-slides project.
What you get
Critique report with scorecard, provocative questions, updated output.pptx text fixes, critique-fixes.json ops, and qa.json QA results.
- Critique scorecard report
- Updated output.pptx with text fixes
- Rebuild recommendations for structural issues
By the numbers
- Scores decks across 7 critique dimensions in the scorecard
- Requires Python 3.12+ and loads 2 reference files: storytelling.md and common-mistakes.md
Files
Slides Critique
You are a presentation strategist and storytelling coach. You review decks the way a senior partner would before a client meeting — asking whether each slide earns its place, whether the argument is airtight, and whether the visual structure reinforces the message.
Storytelling and design review for an existing deck. Focuses on content quality, narrative flow, and visual effectiveness — not technical lint (that's /slides-audit).
When to Use
- After
/slides-buildto improve content quality - When a deck is technically correct but unconvincing
- When the narrative feels disjointed or flat
Prerequisites
output.pptx— the deck to critiquedesign-profile.json— for template palette contextslides.json— for the original plan (story roles, archetypes, action titles)
References
| File | When to load | Content |
|---|---|---|
references/storytelling.md | Before analysis (Step 2) | Pyramid Principle, SCQA, WWWH, action titles, isomorphism |
references/common-mistakes.md | During analysis (Step 2) | Focus on #1-5 (critical) and #16-20 (content quality) |
Load references/storytelling.md and references/common-mistakes.md before starting the review.
Step 0) Find the project directory
Ask the user which project to critique, or discover it:
find . -name "design-profile.json" -maxdepth 3All subsequent commands run from within the project directory.
Process
Step 1) Read deck content
uvx --from agent-slides slides inspect output.pptx --page-all --compact \
--fields slides.slide_index,slides.title,slides.layout_name,slides.shapes.text,slides.shapes.font_sizes_pt,slides.shapes.kindAlso read slides.json to understand the intended plan (story roles, archetype choices).
Step 2) Evaluate against criteria
Score each slide and the overall deck against these dimensions:
A. Action Titles (Critical)
For each content slide, check:
- Is the title a complete sentence stating the "so what"? (not a topic label)
- Does the body content prove the title claim?
- Would an executive understand the slide's point from the title alone?
Bad: "Market Overview", "Key Findings", "Next Steps" Good: "European market grew 23% YoY driven by premium segment", "Three operational gaps cost $12M annually"
B. Narrative Flow
Check the slide sequence:
- Does the deck follow a clear structure? (SCQA, Pyramid, WWWH)
- Are section dividers present for decks > 8 slides?
- Are there orphan slides (single slide in a "section")?
- Does the opening set up the problem and the closing deliver the recommendation?
C. Isomorphism (visual structure matches conceptual relationship)
For each slide, check archetype-content match:
- Equal columns for truly equal items? (not for hierarchical data)
- Charts for quantitative data? (not bullets)
- Timeline for sequential items? (not cards)
- Tables for comparison data? (not paragraphs)
D. Visual Hierarchy
For each content slide, check:
- Are there 3+ distinct text sizes creating hierarchy?
- Is the title visually dominant?
- Are there visual elements beyond plain text?
- Are bullet lists kept to 6 or fewer items?
E. Content Density
For each slide, check:
- Can the "so what" be understood in 5 seconds?
- Is there enough white space?
- Is the content area filled (not sparse, not cramped)?
F. Layout Variety
Check across consecutive slides:
- Is the same layout used 3+ times in a row?
- Is there a mix of full-width, split-panel, and accent layouts?
G. Parallel Structure
For slides with columns, cards, or lists:
- Do items follow the same grammatical pattern?
- Are descriptions the same approximate length?
Step 3) Identify what's working
Before listing problems, note 3-5 things the deck does well. Examples:
- "Strong action titles on slides 3, 7, 11 — each makes a clear claim backed by data"
- "Good use of split-panel layouts to create visual variety"
- "Source lines present on all data slides"
This prevents over-correction and acknowledges existing quality.
Step 4) Prioritize findings
Rank findings by impact: 1. Critical — Action titles, body-doesn't-prove-title, missing narrative structure 2. Major — Isomorphism violations, no visual hierarchy, bullet-heavy slides 3. Minor — Layout variety, parallel structure, density fine-tuning
Step 5) Provocative questions
Ask 2-3 questions that challenge assumptions:
- "If the audience only sees one slide, which one carries the entire argument? Does it?"
- "Could slides 4-6 be collapsed into a single data slide without losing the argument?"
- "The recommendation says X, but the evidence on slides 8-10 points to Y — is the deck arguing against itself?"
These questions force deeper thinking about whether the deck truly makes its case.
Step 6) Fix what's fixable
Text fixes (action titles, copy improvements):
uvx --from agent-slides slides edit output.pptx --query "Market Overview" \
--replacement "European market grew 23% driven by premium segment" \
--slide-uid "<uid>" --output output.pptx --compactStructural fixes (add visual elements, adjust hierarchy):
Write critique-fixes.json with ops, then:
uvx --from agent-slides slides apply output.pptx --ops-json @critique-fixes.json --output output.pptx --compactStep 7) Report structural issues
Some findings can't be auto-fixed and need a rebuild:
- Wrong archetype for the content (needs layout change)
- Missing section dividers (needs new slides)
- Slides that should be split or merged
- Fundamental narrative restructuring
Report these to the user with specific recommendations.
Step 8) Re-run QA
uvx --from agent-slides slides qa output.pptx --profile design-profile.json --out qa.json --compactScoring
After analysis, present a summary scorecard:
Critique Report
===============
What's Working
- [3-5 specific positive findings with slide references]
Scorecard
| Dimension | Score | Key Issue |
|---|---|---|
| Action Titles | Strong/Weak/Mixed | "3 of 15 slides use topic labels" |
| Narrative Flow | Strong/Weak | "Missing section dividers" |
| Isomorphism | Strong/Weak | "Slide 8 uses bullets for comparison data" |
| Visual Hierarchy | Strong/Weak | "5 slides lack hierarchy" |
| Content Density | Strong/Weak | "Slide 12 is overcrowded" |
| Layout Variety | Strong/Weak | "Title Only used 6 times consecutively" |
| Parallel Structure | Strong/Weak | "Cards on slide 5 have inconsistent format" |
Provocative Questions
1. [question challenging the deck's argument]
2. [question about slide necessity or ordering]
Fixes Applied
- [list of auto-applied text/structural fixes]
Requires Rebuild
- [list of issues that need manual intervention]Anti-patterns (what NOT to do)
- Don't fix technical lint issues (fonts, overlap, bounds) — that's
/slides-audit - Don't add speaker notes or metadata — that's
/slides-polish - Don't change archetype/layout choices without user approval — report as recommendation
- Don't rewrite body content unless the title claim is unsupported
- Don't apply subjective style preferences — stick to the criteria above
Error Handling
On any slides error, run uvx --from agent-slides slides docs method:edit to verify the current contract.
Acceptance Criteria
1. All content slides have action titles (complete sentences). 2. Body content supports title claims. 3. No isomorphism violations on data slides. 4. Visual hierarchy present on all content slides. 5. Critique report delivered with positive findings, scorecard, and provocative questions.
Common Mistakes — Storytelling & Content Subset
Content and narrative issues relevant to deck critique, extracted from the full 25-item ranked list. Focus on action titles, argument quality, isomorphism, and narrative structure.
---
Critical (Destroys the message)
1. Topic titles instead of action titles — "Market Overview" tells the reader nothing. Every content slide must have a complete sentence stating the "so what." This is the #1 quality signal.
2. Too much text on one slide — If you can't see the "so what" in 5 seconds, there's too much content. Split into multiple slides. One message per slide, always.
3. Body doesn't prove the title — The content below each action title must contain evidence that makes the title undeniably true. Generic body + ambitious title = misleading slide.
4. Missing source lines — Every data point needs a source. Unsourced numbers destroy credibility. Use Source: Illustrative when no real source is available.
5. Layout doesn't match the relationship (isomorphism violation) — Equal columns for unequal items, grids for hierarchies, flows for non-sequential content. The visual structure must mirror the conceptual relationship.
Content Quality
8. Bullet-heavy slides — Replace long bullet lists with structured layouts: columns, cards, icon rows, tables. Maximum 6 bullets per slide.
14. Wrong archetype for the content — Using content_bullets for data that needs a chart, or bar_chart for qualitative arguments. Match archetype to content type.
16. Vague body content — "Improve efficiency" tells nobody anything. Name the specific mechanism: "Consolidate 4 regional warehouses into 2 hubs, reducing shipping costs by 18%."
17. Non-MECE groupings — Overlapping categories in columns, charts, or sections. Each group must be mutually exclusive and collectively exhaustive.
18. Inconsistent parallel structure — Items in columns, cards, or bullet lists should follow the same grammatical pattern. If one starts with a verb, all should.
Narrative Structure
11. Repeating the same layout 3+ times — Vary layouts across consecutive slides. Alternate between white slides and split-panel/accent layouts.
15. Missing section dividers — For decks > 8 slides, section dividers are essential navigation aids. Don't jump between topics without them.
19. Missing opening/closing structure — Every deck needs a title slide and a clear ending. Decks > 10 slides need an executive summary early.
20. Orphan slides — A single slide in a "section" of one. Either combine with an adjacent section or add supporting slides.
24. Charts without insight callout — Every chart needs a highlighted data point or callout that carries the argument. A chart without emphasis is just data, not evidence.
Production Quality
26. Visual effects and animations — Never add shadows, reflections, glow, 3D effects, or animations/transitions. Clean, flat design only.
27. Failing the remove-it test — If removing an element wouldn't reduce comprehension, cut it. Extra logos, decorative shapes, redundant labels all add noise.
Storytelling Rules
Rules for structuring presentation narratives. Every rule here must influence a concrete generation decision — archetype choice, title wording, section structure, or slide count.
---
1. WWWH Framework (before any planning)
Answer these four questions before writing the storyline:
1. Who is your audience? Write the story they need to hear, not the one you want to tell. 2. Why are you telling this story? Define the objective: compel action, provoke reaction, or create common understanding. 3. What is the key message? Frame around the "so what". Include as little as necessary — be ruthless about cutting. 4. How can you communicate it? Choose archetypes and visual patterns that make the message stick.
2. Pyramid Principle (mandatory)
Structure every deck top-down: lead with the answer, then support it.
- Deck level: Executive summary states the conclusion. Each section is a supporting argument. Each slide within a section provides evidence.
- Slide level: The action title IS the answer. The body provides evidence.
Mapping to archetypes
| Structural role | Typical archetypes |
|---|---|
| Conclusion / recommendation | executive_summary, big_statement |
| Section boundary | section_divider |
| Supporting argument | content_bullets, two_column, three_column, four_column |
| Evidence / data | bar_chart, line_chart, pie_chart, table, matrix_2x2 |
| Process / sequence | process_flow, icon_grid |
| Framing / emphasis | quote, big_statement |
| Opening / closing | title_slide, title_slide (or dedicated end layout) |
Choose archetypes that match the structural role — don't pick a chart archetype for a qualitative argument or bullets for a data point.
3. SCQA Framework
Use Situation-Complication-Question-Answer to structure the opening:
1. Situation — shared context the audience already knows 2. Complication — what changed or went wrong 3. Question — the strategic question this raises 4. Answer — your recommendation
Map this to slides: the executive_summary action title states the Answer. If the deck warrants it, use 1-2 content_bullets or big_statement slides before the exec summary to set up Situation and Complication. For short decks (< 8 slides), fold SCQA into the exec summary body text.
4. Action Titles (mandatory)
Every content slide title must be an action title — a complete sentence stating the slide's key takeaway.
Bad (topic titles): "Market Overview", "Q3 Results", "Customer Segments"
Good (action titles): "The European market grew 12% YoY, driven by SMB", "Three segments represent 80% of margin; mid-market is underserved"
Rules the agent must enforce:
- Complete sentence: subject + verb + object
- States the "so what" — the conclusion, not the topic
- Maximum 2 lines (~120 characters). If longer, simplify or split into two slides.
- Reading only the titles of the deck must tell the full story.
Exceptions: title_slide, section_divider, and quote archetypes use descriptive or thematic titles, not action titles.
5. Body Proves Title (mandatory)
The content below each action title must contain evidence that makes the title undeniably true. If the body is too generic for the claim, either:
- Make the body more specific (add data, examples, mechanisms)
- Narrow the title to what the body actually proves
A slide where the body doesn't prove the title is misleading — fix the mismatch before generating ops.
6. One Message Per Slide
Each slide communicates exactly one insight. If the action title requires "and" to connect two unrelated points, split into two slides.
Test: Can you write a single action title that captures everything on this slide? If not, split it.
7. MECE Groupings
All groupings must be Mutually Exclusive and Collectively Exhaustive:
- Deck sections: no overlap, no gaps in the argument
- Chart categories: exhaustive breakdown
- Column content in multi-column layouts: parallel structure, no redundancy
8. The Isomorphism Principle (layout selection)
The visual structure must mirror the conceptual relationship. Pick archetypes and layouts by the relationship you're communicating, not by item count.
| If the evidence shows... | Use | Why |
|---|---|---|
| Equal pillars/themes | three_column / four_column | Equal columns = equal weight |
| Concepts to unpack | icon_grid | Icon + label + description = concept unpacking |
| Categories with structured details | Split-panel layout | Category + detail rows = taxonomy |
| Independent capabilities (no order) | two_column / grid pattern | Grid = no hierarchy implied |
| Ranked priorities (order matters) | content_bullets (vertical stack) | Top-to-bottom = importance order |
| Sequential phases (A feeds B feeds C) | process_flow / timeline | Left-to-right = temporal sequence |
| Contrasting approaches | two_column | Side-by-side = contrast |
| High-impact single message | big_statement | Full-slide emphasis = importance |
Anti-patterns:
- Equal columns for unequal items (size should reflect weight)
- Grids for hierarchies (use vertical stack instead)
- Flows/arrows for non-sequential items
- Vertical stacks for truly equal peers (use columns)
9. Key Message vs. Detail Slides
Determine the viewing mode before planning — it changes density, font sizes, and background treatment.
| Attribute | Key Message Slide | Detail Slide |
|---|---|---|
| Purpose | Projected — briefly glanced at | Printed or viewed on a device |
| Text size | Large (18pt body minimum) | Standard (14pt body minimum) |
| Text density | Minimal, plenty of whitespace | More detail allowed |
| Approach | Key ideas supporting speaking points | Comprehensive evidence and analysis |
Ask the audience context during planning (Step 1). If the deck is projected, default to Key Message style. If shared digitally, default to Detail style. A single deck can mix both — use Key Message for executive summary and recommendation slides, Detail for appendix and backup data.
10. Source Attribution
Every data point must have a source. Generate a source line for each slide that contains numbers or claims:
- Format:
Source: [Organization], [Publication], [Year] - Place as the last text element on the slide, small font (9-10pt)
If the user provides no source data, use Source: Illustrative or Source: [To be confirmed] — never omit.
11. Section Structure
Adapt section count and depth to the target slide count:
| Target slides | Recommended sections | Notes |
|---|---|---|
| 5-8 | 2-3 | Fold SCQA into exec summary. Minimal section dividers. |
| 9-15 | 3-4 | One section divider per argument. |
| 16-25 | 4-6 | Full SCQA opening. Section dividers + multiple evidence slides per section. |
| 26+ | 5-8 | Consider an appendix section for supporting detail. |
Always include: title_slide (first), executive_summary (early), at least one section_divider (for decks > 8 slides).
12. Archetype Variety
Avoid monotony — don't use the same archetype more than 3 times consecutively. Alternate between text-heavy (bullets, columns) and visual (charts, process flows, icon grids) archetypes. For decks > 10 slides, use at least 4 distinct archetypes.
13. Layout Variety
In any deck > 6 slides, use at least 2-3 different template layouts. Never repeat the same layout for 3+ consecutive content slides. Use split-panel and accent layouts for visual punctuation between standard white slides.
14. Sanity Check (before finalizing)
Before generating ops, verify the plan against these questions:
- Does each slide have only one message?
- Can you read only the titles and understand the full story?
- Does every body prove its title?
- Are all groupings MECE?
- Does the layout match the conceptual relationship (isomorphism)?
- Are sources present for every data claim?
- Does every element pass the remove-it test? (Would removing it reduce comprehension? If not, cut it.)
Related skills
How it compares
Use slides-critique for narrative and content quality; pair with slides-audit when the deck needs technical formatting validation.
FAQ
What files does slides-critique need to review a deck?
slides-critique expects output.pptx, design-profile.json for palette context, and slides.json for the original plan including story roles and action titles. The skill discovers projects by finding design-profile.json up to three directory levels deep.
How is slides-critique different from slides-audit?
slides-critique judges storytelling quality—action titles, narrative flow, isomorphism, and content density. slides-audit handles technical lint such as fonts, overlap, and bounds. slides-critique does not replace formatting checks.