| name | presentation-builder |
| description | End-to-end presentation creation through structured phases: research, requirements, design, visual generation, implementation, and iteration. Handles PPTX, HTML, PDF, and DOCX output formats via document-skills. Use this skill whenever the user wants to create a presentation, slide deck, pitch deck, talk, or lecture — even if they just say "build me a deck about X" or "I need slides for a talk on Y." Also use when the user has existing content (notes, docs, research) they want turned into a presentation. Triggers on: "create presentation", "build slides", "make a deck", "presentation about", "slide deck for", "put together a talk", "pitch deck", "build me a pptx", or any request that involves creating structured visual content for an audience.
|
| argument-hint | <topic or description of the presentation> |
Presentation Builder
Build professional presentations through a structured, iterative workflow. This skill captures the
methodology of spending the most effort on requirements and design (shift-left), using AI to handle
orchestration, and verifying output through multi-agent review.
First Run: Prerequisites & Setup
On first invocation, check and install dependencies in order. Read references/setup.md for the
full setup process including MCP configuration.
1. document-skills plugin (required for output format skills + /generate-image):
ls ~/.claude/plugins/cache/anthropic-agent-skills/document-skills/ 2>/dev/null
If not found, guide the user:
"This skill needs the document-skills plugin for building presentations. Let me set that up."
Then run:
/plugin marketplace add anthropics/skills
/plugin install document-skills@anthropic-agent-skills
/reload-plugins
2. /generate-image skill (required for AI image generation):
Image generation is delegated to the /generate-image skill. If the user's environment does not have it set up, guide them:
"Image generation is handled by the /generate-image skill. Run /generate-image:setup to configure it — you'll need a free Replicate API token from replicate.com/account/api-tokens. The setup skill handles MCP configuration."
Read references/setup.md for the delegation details and Q8 fallback choices.
3. Review tools (optional, enhances quality):
- Check for
/bmad-party-mode (Tier 1 review)
- Check for
/reflexion:critique (Tier 2 review)
- If neither available, the skill uses built-in multi-perspective review prompts (Tier 3)
After first successful setup, create ~/.claude/.presentation-builder-setup-complete
so subsequent runs skip the check. This is global (not per-project) because the prerequisites
(document-skills, /generate-image) are machine-level, not project-level.
Complexity Detection
Before starting, assess presentation complexity to choose the right workflow:
Quick Path (<10 slides, single topic, no external research needed):
- Skip Phase 1 (Research)
- Compress Phases 2-3 into one brainstorm
- Skip Phase 5 (Design Review)
- Generate visuals and build in one pass
- Example: "Make a 5-slide team update on Q4 progress"
Standard Path (10-20 slides, moderate complexity):
- Full workflow with single review passes
- Example: "Build a 20-minute talk on our new API for the dev team"
Complex Path (>20 slides, multiple sources, mixed audience):
- Full workflow with multi-agent reviews at every gate
- Example: "Put together a 45-minute presentation on advanced AI coding practices"
Workflow Overview
The workflow has three user-facing stages with internal phases. At each check-in, show the
progress indicator so the user always knows where they are.
STAGE 0: TEMPLATE (conditional — only when user provides a branded PPTX)
Phase 0: Template Analysis ───→ template-analysis.md + template-assets/
STAGE 1: DISCOVERY
Phase 1: Research & Gather ──→ MATERIALS_INDEX.md
Phase 2: Requirements ────────→ Presentation design spec
STAGE 2: DESIGN
Phase 3: Multi-Agent Review ──→ Refined spec
Phase 4: Style Guide ─────────→ style-guide.md
Phase 5: Design Review ───────→ Approved spec + style guide
Phase 6: Visual Generation ───→ AI-generated images
STAGE 3: BUILD
Phase 7: Implementation ──────→ Build scripts (modular)
Phase 8: Code Review ─────────→ Verified scripts
Phase 9: Build & Iterate ─────→ Final presentation file
Check-in format: At the end of each phase, present:
"[Stage N/3 — Phase M: Phase Name] Here's what we have so far: [brief summary].
Does this look right? Anything to adjust before we move on?"
Do NOT skip phases -- the shift-left principle means most effort goes into Stages 1-2.
The implementation in Stage 3 becomes almost mechanical because the design is solid.
Resuming a Presentation
If invoked in a directory with existing presentation artifacts, detect partial progress:
| If this exists... | Resume from... |
|---|
template-analysis.md only | Phase 1 (Research) or Phase 2 (Requirements) |
MATERIALS_INDEX.md only | Phase 2 (Requirements) |
Design spec in docs/superpowers/specs/ | Phase 3 (Review) |
style-guide.md | Phase 5 (Design Review) or Phase 6 (Visuals) |
images/ directory with files | Phase 7 (Implementation) |
build-deck/ directory with scripts | Phase 8 (Code Review) or Phase 9 (Build) |
.pptx/.html/.pdf/.docx output | Phase 9 (Iterate) |
Present what was found and ask: "It looks like you have a presentation in progress. Want to
pick up where you left off, or start fresh?"
Phase Details
Phase 0: Template Analysis (conditional)
Read references/phase-0-template.md for the detailed process.
Summary: When the user provides an existing branded PPTX template, extract colors, fonts,
and media assets from it. Unzip the PPTX, parse ppt/theme/theme1.xml for the color scheme
and font families, copy ppt/media/* to template-assets/, rename assets to semantic names,
and generate template-analysis.md at the project root.
When to run: Ask during initial setup whether the user has a branded PPTX template. If yes,
run Phase 0 before Phase 1. If no template, skip entirely.
Outputs: template-analysis.md + template-assets/ directory with renamed assets.
Phase 1: Research & Gather
Read references/phase-1-research.md for the detailed process.
Summary: Identify and collect source material. Use parallel agents to explore multiple
locations simultaneously. Organize into MATERIALS_INDEX.md.
If the user has NO source materials (starting from scratch), skip this phase and go directly
to Phase 2. The brainstorm will generate content.
Outputs: gathered-materials/ directory + MATERIALS_INDEX.md
First: Create a project directory if one doesn't exist. Ask the user where to create it
or use the current working directory.
Phase 2: Requirements & Design
Read references/phase-2-requirements.md for the detailed process.
Summary: Brainstorm structure through one-question-at-a-time dialogue. Cover: primary
takeaway, audience, duration, sections, timing, demos, output format.
Outputs: Design spec at docs/superpowers/specs/YYYY-MM-DD-<topic>-design.md
Phase 3: Multi-Agent Review
Read references/review-tiers.md for the tiered review detection process and
references/fallback-review-prompts.md for the Tier 3 persona prompts,
substance expectations, and substantiated-waiver format.
Refine the design spec through multi-perspective review using the best available mechanism.
Phase-complete gate: Phase 3 is NOT complete until:
- The design spec contains a "Review Log (Phase 3)" section with ≥ 3 substantive findings per persona (or a substantiated waiver per persona — bare "no issues from this perspective" does NOT satisfy this gate).
- Every Must-fix finding has a corresponding spec edit or explicit documented dismissal in the review log.
A common Sonnet failure pattern is producing shallow reviews (one or zero findings per persona) or advancing without applying Must-fix edits. Do NOT advance to Phase 4 in that state.
Outputs: Refined design spec with Review Log (Phase 3) section.
Phase 4: Style Guide
Read references/phase-4-style-guide.md for the detailed process.
Summary: Create style-guide.md governing ALL visual output -- slides and generated images.
Present to user for approval before generating any images.
Outputs: style-guide.md
Phase 5: Design Review
Run another review pass, this time on the complete design (spec + style guide together) using the same tiered mechanism from Phase 3.
For Complex presentations, use full multi-agent review. For Quick/Standard, an inline check is sufficient — but substance expectations and the substantiated-waiver rule still apply.
Phase-complete gate: Phase 5 is NOT complete until:
- The spec contains a "Review Log (Phase 5)" section with ≥ 3 substantive findings per persona (or substantiated waiver), reviewing spec + style-guide together.
- Every Must-fix finding has a corresponding spec/style-guide edit or explicit dismissal.
Outputs: Final approved spec + style guide with Review Log (Phase 5) section.
Phase 6: Visual Generation
Hard prerequisite: /generate-image skill configured and working AND Q8 ≠ text-only.
If /generate-image is absent, STOP (Phase 2 should have caught this — see setup re-entry path in references/phase-2-requirements.md).
If Q8 = text-only, skip Phase 6 entirely and advance to Phase 7.
Read references/phase-6-visuals.md and follow it end-to-end. Delegate all image generation to the /generate-image skill — do not call mcp__replicate__* tools directly from this skill. This phase is not complete until images/ contains one file per image in the style guide's plan. Writing IMAGE_PLAN.md without files on disk is a known failure mode — do not advance to Phase 7 in that state.
Important: This phase runs BEFORE implementation because build scripts reference image paths.
Outputs: images/ directory with all generated assets (unless Q8 = text-only).
Phase 7: Implementation Scripts
Read references/phase-7-implementation.md for the detailed process.
Summary: Build the presentation using modular scripts. Delegate to the
document-skills skill for the chosen format:
- PPTX:
/pptx skill (pptxgenjs). Fallback: if /pptx is
unavailable, use pptxgenjs directly with the modular theme.js +
slides-s{N}.js + build.js architecture. Do NOT stall — this is a
documented fall-through.
- HTML:
/frontend-design skill. Fallback: plain HTML+CSS+JS.
- PDF:
/pdf skill.
- DOCX:
/docx skill.
For PPTX: theme.js + slides-s{N}.js per section + build.js orchestrator.
Speaker notes are NOT optional. Every slide gets the full treatment. Read
references/presentation-best-practices.md for the speaker notes format.
Phase-complete gate: Phase 7 is not complete until build.js runs with
zero errors and produces a presentation file of the expected slide count.
See references/phase-7-implementation.md for details.
Outputs: Build scripts in build-deck/ directory + generated presentation file.
Phase 8: Code Review
Read references/phase-8-code-review.md for the detailed process, including the MANDATORY pre-review build run.
Summary: Run the build FIRST (capture exit code, stderr/stdout, file size). Then review build scripts via the tiered review mechanism (see references/review-tiers.md). Apply Must-fix findings.
Phase-complete gate: Phase 8 is NOT complete until:
- Build was executed with captured output.
- Latest build run exits with code 0.
code-review.md exists at project root with "Review Log (Phase 8)" section containing ≥ 3 substantive findings per persona (or substantiated waivers).
- Every Must-fix finding has a code edit applied or explicit dismissal.
A common Sonnet failure pattern is reviewing code without running the build. Do NOT advance to Phase 9 in that state.
Outputs: Verified build scripts + code-review.md with Review Log (Phase 8).
Phase 9: Build & Iterate
Read references/phase-9-build.md for the detailed process, including post-build PPTX
validation (media count, format verification, contrast check, speaker notes check).
Summary: Run the build, then validate the generated file structurally. For PPTX, unzip
and inspect: verify media count matches the image plan, check for 0-byte files, validate
image formats via magic bytes (catches WebP-as-PNG), and verify speaker notes exist on every
slide. Run contrast validation on all text/background combinations.
Phase-complete gate: Phase 9 is NOT complete until:
- The output file exists at the expected path (e.g.,
output.pptx).
- The output file size exceeds the minimum threshold:
- PPTX: >= 10 KB
- PDF: >= 5 KB
- DOCX: >= 5 KB
- HTML: >= 5 KB
- The slide count matches the spec's total slide count.
- The build invocation is recorded in the transcript (exit code captured).
- (PPTX only) Post-build validation passed — no 0-byte media, no format mismatches,
no contrast ratio below 3:1.
- Any Must-fix validation failures have been resolved and the build re-run.
NOTE: Phase 7's build gate partially overlaps this one — that is
intentional. Phase 7 requires the build was made to work during
implementation; Phase 9 re-verifies the final build after any code-review
edits. Belt-and-suspenders redundancy is deliberate for Sonnet reliability.
A common Sonnet failure pattern is declaring Phase 9 complete based on
build scripts existing rather than on actually running them. Do NOT skip
the build invocation — the build script must execute and produce a file
on disk that passes the size, slide-count, AND structural validation checks.
Iterate based on user feedback using the iteration protocol below.
Outputs: Final validated presentation file.
Iteration Protocol
The modular architecture makes changes cheap:
- Content change: Edit specific
slides-s{N}.js, rebuild
- Image change: Regenerate specific image, update reference, rebuild
- Style change: Update
style-guide.md + theme.js, regenerate affected images, rebuild
- Structure change: Update spec, add/remove section files, update
build.js, rebuild
- Speaker notes: Update notes in specific section file, rebuild
Always rebuild and verify after changes.
Output Format Selection
| Format | Skill | Best For |
|---|
| PPTX | /pptx | Conference talks, corporate presentations, speaker notes |
| HTML | /frontend-design | Interactive presentations, web-based delivery |
| PDF | /pdf | Handouts, read-along decks, print |
| DOCX | /docx | Narrative presentations, proposal decks with heavy text |
Default to PPTX unless the user specifies otherwise.