kai-export-ppt-lite

name	kai-export-ppt-lite
description	Use when exporting HTML presentations to editable PPTX in sandbox environments where Playwright cannot be installed. Pure Python: bs4 + python-pptx + Pillow. Keywords: export pptx sandbox, html to pptx no browser, editable powerpoint from html.

kai-export-ppt-lite

Export HTML presentations to editable PPTX without a browser. Pure Python — no Playwright, no Chrome, no Node.js.

Commands

Command	What it does
python3 <skill-path>/scripts/export-sandbox-pptx.py <file.html> [output.pptx] [--width 1440] [--height 900]	Canonical entrypoint; works as a real file or as inline-retrieved script
python3 <skill-path>/scripts/run-skill-export.py <file.html> [output.pptx] [--width 1440] [--height 900]	Optional tiny bootstrap when the runtime exposes a real installed skill directory
python3 <skill-path>/scripts/run-skill-evals.py --runner fixture	Agent-neutral fixture eval; does not require Codex, Claude, Qoder, network, model auth, or any live agent

Dependencies

Package	Install
beautifulsoup4	Usually auto-installed by skill metadata or pip install beautifulsoup4
lxml	Usually auto-installed by skill metadata or pip install lxml
python-pptx	Usually auto-installed by skill metadata or pip install python-pptx
Pillow	Usually auto-installed by skill metadata or pip install Pillow

Usage

python3 <skill-path>/scripts/export-sandbox-pptx.py <file.html> [output.pptx] [--width 1440] [--height 900]

Local repo execution remains available:

python3 scripts/export-sandbox-pptx.py <file.html> [output.pptx] [--width 1440] [--height 900]

Agent-Neutral Evals

The eval runner is intentionally agent-neutral. Default evaluation uses checked-in normalized fixture traces plus real local PPTX artifact checks:

python3 scripts/run-skill-evals.py --runner fixture --format json --json-out .tmp/skill-evals.json
python3 scripts/test-skill-evals.py

Do not require Codex, Claude, Qoder, network access, model auth, or a live agent environment for the default eval path. Future live runner adapters must be added outside the scorer and backed by real captured trace fixtures.

Sandbox Execution

Do not assume anything about the agent runtime. This skill must tolerate:

• no stable skill root path
• no __file__
• notebook / cell-style inline execution
• dependencies either preinstalled or missing

When this skill runs inside a hosted sandbox, use this decision order:

1. Default path: retrieve and execute scripts/export-sandbox-pptx.py.
2. If the runtime exposes a real installed skill directory, you may instead call:

   python3 <skill-path>/scripts/export-sandbox-pptx.py input.html output.pptx
   

3. If the runtime only supports inline Python execution, retrieve scripts/export-sandbox-pptx.py and inline it directly.
4. Pass the HTML input via the sandbox VFS/attachment mechanism and write the PPTX to the sandbox artifact directory.
5. Let the script self-bootstrap dependencies first. Skill metadata / requirements.txt are optimizations, not assumptions.
6. scripts/run-skill-export.py is an optional optimization only. Do not require it for correctness.

Why this is the stable path

• scripts/export-sandbox-pptx.py is the only execution surface that does not require assuming a skill root, env var, or installer-specific layout.
• The exporter now tolerates missing __file__, degrades when vendored contracts/ are absent, and attempts runtime dependency bootstrap before failing.
• run-skill-export.py remains useful when the environment is richer, but it is not part of the correctness boundary.
• If vendored contracts/ are unavailable, export still runs; only preset-specific fidelity hints degrade.

Prohibited patterns

• assuming an installed skill directory exists
• assuming __file__ exists in notebook / cell-based sandboxes
• assuming build-time dependency installation always happened
• making run-skill-export.py the only supported path

Options

Flag	Description
--width N	Slide width in pixels (default: 1440)
--height N	Slide height in pixels (default: 900)
--with-chrome	Add page counter and nav dots

What it extracts

HTML Element	PPTX Output
h1-h6, p, li, span	Editable text boxes
div with background/border	Rounded rectangles with fill
table, tr, td, th	Cell rectangles with text frames
img (http/file/data-uri)	Embedded pictures
svg	Placeholder rectangle

Limitations (no browser)

• CSS variables: only :root static substitution
• Layout: simulated flex-column, not pixel-perfect
• Computed styles: unavailable, relies on inline style + <style> blocks
• JavaScript: not executed, so IntersectionObserver-gated content won't appear

Comparison

	kai-html-export (full)	kai-export-ppt-lite
Runtime	Playwright + Chrome	Pure Python
Sandbox compatible	❌	✅
Output quality	Pixel-perfect	Semantic layout
Editable text	✅	✅
Image mode PPTX	✅ (screenshot)	❌
Dependencies	playwright, python-pptx	bs4, lxml, python-pptx, Pillow