视觉画布设计
canvas-design
基于设计原则创作静态视觉作品,用于海报与品牌图形。
pptx-from-layouts【文档与办公】Generate and edit consultant-grade PowerPoint decks from markdown outlines using a template's real slide-master layouts (not text overlays). Use when the user wants to create slides/a deck/a presentation from an outline or brief, modify an existing .pptx, turn their own PowerPoint into a reusable template, or check a deck's quality. Triggers: "make a…
独立 GitHub 快照 · 自动分类 · 以原作者文档为准
Generate consultant-ready PowerPoint presentations from markdown outlines by filling the template's actual slide-master layouts. Core thesis: use the template's layouts and placeholders — never overlay text boxes on slides.
This skill is Step 3 of a three-skill pipeline. Each step is its own skill:
STEP 1 — PROFILE STEP 2 — AUTHOR STEP 3 — RENDER ◀ you are here
pptx-profile pptx-author pptx-from-layouts
─────────────── ─────────────── ─────────────────
your-template.pptx slides.md (+ [HINT:]) slides.md + config
│ │ │
▼ ▼ ▼
catalog.py ──catalog.md──▶ write & lint ──slides──▶ generate.py ─▶ deck.pptx
└────────────────── config.json ───────────────────▲ (validated)
"what layouts do I "write slides that "fill the template's
have?" name real layouts" actual placeholders"
pptx-profile: catalog.py turns a .pptx into a layout catalog
(the menu of [HINT:] names) + a render config.json. One-time per template.pptx-author: write slides.md with [HINT: layout_name] per
slide, then lint_hints.py validates the hints against the catalog.generate.py fills the named layouts and validates.You can also start here directly with the bundled Inner Chapter template (no Step 1 needed) — it's the default and the running example throughout this doc.
[HINT:] — author against real layoutsA slide can name the exact template layout it should render on:
# Slide 2: Three outcomes the research must achieve
[HINT: column-3-centered-a]
[Column 1: Level-set] …
[HINT:] resolves against the active template config (from Step 1), so it
works for any profiled template — not just Inner Chapter. It's equivalent to
the **Layout:** marker, and takes precedence over **Visual:**. Use [HINT:]
when you know the layout (precise); use **Visual:** when you want the engine to
auto-pick one for a semantic intent (e.g. process-4-phase). Mismatched hints
fall back to auto-detection with a warning — so lint in Step 2 first.
Requires Python 3.10+, python-pptx, and pydantic:
pip install python-pptx pydantic # or: pip install -r requirements.txt
If you hit ModuleNotFoundError: No module named 'pptx' or 'pydantic', this
is why. The entry-point scripts set their own PYTHONPATH; run them directly
with python.
| The user wants… | Do this |
|---|---|
| A deck from a brief / notes / source doc (no outline yet) | Delegate to pptx-outline-architect, then generate |
| A deck and they already wrote an outline | generate.py → validate |
| To use their own PowerPoint as the template | Delegate to pptx-template-onboarder (one-time setup) |
| Small text fixes / reorder on an existing deck (<30% slides) | edit.py (inventory → replace / reorder) |
| Layout changes, added/removed slides, or >30% churn | Regenerate via generate.py — never edit |
| To know if a deck is good | Delegate to pptx-deck-qa or run validate.py |
**Visual: <type>** declaration on EVERY
content slide. Choose visual types with the decision order below — do not
default everything to bullets.python scripts/generate.py outline.md -o deck.pptx --validate
(Add --template <t.pptx> --config <c.json> for a custom template.)**Visual:** declaration.generate.py exited successfully and the .pptx exists.validate.py reports zero errors and no text overflow.If any box is unchecked, the task is not finished — fix it or say explicitly what blocked you.
Canonical definitions in agents/; install to ~/.claude/agents/ to use them.
| Subagent | Use it to |
|---|---|
pptx-outline-architect | Turn raw material into a generation-ready outline.md with correct visual types |
pptx-template-onboarder | Onboard a user's .pptx as a reusable, on-brand template (one-time) |
pptx-deck-qa | Validate a deck and apply a surgical fix or recommend regeneration |
They chain as architect → generate.py → QA; for your own template,
onboarder runs once up front. (Full write-up: docs/workflows.md in the repo.)
# Generate from outline (bundled Inner Chapter template, with validation)
python scripts/generate.py outline.md -o deck.pptx --validate
# Use your own template
python scripts/generate.py outline.md -o deck.pptx \
--template your-template.pptx --config your-template-config.json --validate
# Edit existing deck (dump inventory, edit its JSON, apply as a file)
python scripts/edit.py deck.pptx --inventory -o inv.json
# edit inv.json: change the text on the paragraph(s) you care about
python scripts/edit.py deck.pptx --replace inv.json -o edited.pptx
# Validate quality
python scripts/validate.py deck.pptx
# Profile a custom template (one-time onboarding)
python scripts/profile.py template.pptx --name my-template --generate-config
The bundled template is only a demo. To make on-brand decks, onboard your own
.pptx once — profile it, generate its layout-mapping config, smoke-test,
and reuse it for every future deck.
python scripts/profile.py /path/to/your-template.pptx \
--name your-template --generate-config --output-dir templates/
This emits your-template-config.json. After that, every deck is a single
generate.py call with --template/--config. Full guide:
rules/bring-your-own-template.md. To automate the whole thing, delegate to
pptx-template-onboarder.
Declare in outlines with **Visual: type-name**.
| Type | Use When |
|---|---|
process-N-phase | Sequential steps (N=2-5) |
comparison-N | Side-by-side options (N=2-5) |
cards-N | Non-sequential parallel items (N=2-5) |
data-contrast | Two opposing metrics |
quote-hero | Powerful quote |
hero-statement | Single punchy statement ONLY |
timeline-horizontal | Date-based sequences |
table | Genuinely tabular data |
bullets | Default (3+ items) — last resort |
Decision order: sequence → comparison → parallel items → data contrast →
quote → table → hero → bullets. Full reference + per-type length limits:
rules/visual-types.md.
| Inline | Result | Paragraph | Result | |
|---|---|---|---|---|
{blue}text{/blue} | IC brand blue | {bullet:-} | Dash bullet (–) | |
{bold}text{/bold} | Bold | {bullet:1} | Numbered | |
{italic}text{/italic} | Italic | {level:N} | Indent level | |
{question}text?{/question} | Blue italic | |||
{signpost}LABEL{/signpost} | Section label |
# Project Overview
**Visual: hero-statement**
Transforming operations through digital innovation
# Our Approach
**Visual: process-4-phase**
[Column 1: Discover]
- Stakeholder interviews
- Competitive audit
[Image: research process]
[Column 2: Define]
- Workshop facilitation
- Strategic framework
[Column 3: Design]
- Solution architecture
- Prototype development
[Column 4: Deliver]
- Implementation
- Training & handover
# 1. Dump the inventory — this is the schema replace.py consumes
python scripts/edit.py project.pptx --inventory -o inv.json
# 2. Open inv.json, find the target paragraph, change only its "text" field.
# e.g. inv.json["slide-2"]["shape-3"]["paragraphs"][0]["text"] = "Q2 2026"
# 3. Apply, then validate
python scripts/edit.py project.pptx --replace inv.json -o updated.pptx
python scripts/validate.py updated.pptx
| Change Type | Action |
|---|---|
| New presentation | generate.py |
| Typos/values (< 30% slides) | edit.py |
| Reorder slides | edit.py --reorder |
| Layout changes | Regenerate |
| Add/remove slides | Regenerate |
| > 30% slide changes | Regenerate |
**Visual:** declaration — the parser default may not match.hero-statement for 3+ items or multi-sentence content.process-N-phase).comparison-N).Author-facing rules in rules/:
outline-format.md, visual-types.md, typography.md, columns.md,
tables.md, editing.md, bring-your-own-template.md, decisions.md.
Reference: references/layouts.md (Inner Chapter layout indices).
Subagents: agents/ (this skill). Workflows write-up: docs/workflows.md (repo).