HTML Slide to PPTX Converter
Turns structured single-slide or small-deck HTML files into natively editable PowerPoint slides instead of screenshot-based image exports.
Evidence shows a purely local file-conversion tool with only cheerio and pptxgenjs as dependencies; no signs of network calls, sensitive-data handling, or external side effects; check-env self-check, explicit install steps, and intermediate model dumps give reasonable data-flow transparency. Deducted for: core script sources not provided in this review scope, so least-privilege and rollback behavior cannot be verified; conflicting license metadata (ISC in package vs Apache-2.0 at repo level) leaves attribution incomplete.
Docs are self-consistent with a clear workflow (identify preset → convert → QA → iterate), plus environment self-check, dependency troubleshooting for missing cheerio/pptxgenjs, portability notes, and a preflight QA layer with decent failure feedback. Deducted for: no independent test suite (the test script only points at check-env); key paths are not reproduced, and static calibration caps this at 10.
Scope and non-goals (not a universal HTML/CSS converter, not a screenshot exporter) are explicit; preset decision rules are thorough; semantic trigger conditions (user has structured HTML slide needing editable PPTX) are clear; qa-heuristics.md is in Chinese, so Chinese support is good; dependencies come only from npm with no entirely unreachable overseas service. Deducted for: only two very narrow preset families are supported, and trigger precision is unverified without execution.
Documentation is well-layered (SKILL.md → README → references) with preset templates, decision rules, QA heuristics, and roadmap, giving a clear extension path. Deducted for: conflicting license metadata (ISC vs Apache-2.0), empty author field, no versioning/changelog, and no explicit maintenance ownership; roadmap items are mixed with delivered capability.
The core value claim (structured HTML → editable native PPTX objects instead of screenshots) is clear, marginal value is plausible for analyst-style fixed layouts, and the workflow includes QA and iteration. Deducted for: no verified representative outputs; editability and fidelity are entirely unsupported by execution evidence, and thin evidence lowers the score below the static cap of 7.
A lockfile with integrity hashes, detailed engine limitations citing PptxGenJS/python-pptx issue numbers, and prototype source-file paths constitute auditable primary material. Deducted for: all key claims (conversion quality, QA effectiveness) lack execution or third-party corroboration; cited issues cannot be verified from source files; coverage is thin.
- This is a static review; no scripts were executed. Conversion quality, editability, and QA effectiveness are unverified—test on your own samples before use.
- License metadata conflict: the package declares ISC while the repository README declares Apache-2.0; confirm the actual license before commercial use.
- Only two very narrow preset families (v9-architecture, ai-runtime-page) are supported; structurally different HTML requires writing a new preset at non-trivial cost.
- No independent test suite is provided; the test script only runs an environment check, so reliability rests mainly on documentation.
- Publisher identity is unverified by the FollowSkills registry; maintenance and update paths are unclear.
What does this skill do, and when should you use it?
This is a preset-driven conversion tool: it reads structured HTML slides, parses semantic blocks, and maps them to native PowerPoint objects such as text boxes, shapes, chips, arrows, and panels. It only promises fidelity for supported slide families, not arbitrary HTML/CSS; for a new layout family a new preset must be added first. Every generation runs a preflight QA pass, so fidelity is approached iteratively. The code lives in the html-slide-to-pptx/ directory of the mucsbr/ppt-agent-workflow-san repository and is licensed under Apache 2.0.
Reads an input HTML slide file, determines whether it matches a supported slide family (preset), parses it into a small intermediate model, and generates a PPTX with native text and shapes via a bundled script. It provides npm run check-env for environment checks, node scripts/html_to_pptx.js <input.html> <output.pptx> for conversion (with optional --preset and --dump-model flags), and node scripts/preflight_qa.js <model.> for QA. If no preset fits, you add a new one by implementing three layers per the docs: DOM extraction rules, layout/render mapping rules, and QA/preflight rules.
- Users producing Gartner/analyst-style technical slides who want an editable PPTX from an HTML design draft rather than a screenshot.
- Technical staff converting architecture pages (layered architecture stack, driver panels, judgement chain panels) into editable PowerPoint.
- Product or solutions teams moving AI runtime pages (input/output chip rows, modular runtime box, support layer cards) into PPT.
- Teams running the full documented workflow: generate HTML/PNG previews first, confirm layout, then convert to editable PPT.
- Engineering-minded users who can read and extend preset rules to support new fixed-layout HTML page families.
What are this skill's strengths and limitations?
- Produces natively editable PowerPoint objects rather than screenshot embeds, so slides stay editable afterwards.
- Built-in preflight QA checks every generated result for spacing, overflow, and arrow direction issues.
- Clear preset discipline: fidelity is promised only for supported layouts, avoiding unrealistic arbitrary-HTML claims.
- Comprehensive extension docs (setup, preset decision rules, preset template, QA heuristics, roadmap) make it extendable by engineering users.
- Only two presets exist (v9-architecture and ai-runtime-page); HTML that does not match cannot be converted directly.
- Explicitly does not promise arbitrary HTML/CSS fidelity; a genuinely new information architecture requires developer work to implement a three-layer preset.
- No repo topics, test suite, or adoption evidence; documentation is inconsistent (the README says two subprojects while the repo description says three bundled skills).
- Requires Node.js and npm dependencies, and treats browser rendering as reference only, never as the generation engine.
How do you install this skill?
Clone the repo: git clone https://github.com/mucsbr/ppt-agent-workflow-san.git. The skill lives in the html-slide-to-pptx/ directory (containing SKILL.md). On a new machine or fresh agent, first read references/setup.md inside that directory, then run npm run check-env in the skill directory; if dependencies are missing, run npm ci or npm install. The exact placement path for your agent client is not documented in the source.
How do you use this skill?
- Give your HTML slide to the agent and have it check whether it matches an existing preset (v9-architecture or ai-runtime-page). 2. Convert: node scripts/html_to_pptx.js input.html output.pptx --preset=v9-architecture (optionally --dump-model model.). 3. Run preflight QA: node scripts/preflight_qa.js model. --preset=v9-architecture --report report.. 4. Open the PPTX and inspect spacing, overflow, and arrow direction issues; iterate the preset mapping if fidelity falls short. If the structure matches no preset, add a new preset first per SKILL.md, then convert.