DocuMint Interactive HTML Skill Plugin v0.3.2
One familiar UI, no DocuMint CLI. This Skill bundles a complete HTML shell exported from the actual DocuMint VS Code Local HTML CSS, JavaScript and project-map markup. It uses a project JSON object to render an offline HTML artifact in a browser. No MCP server, VPS, API key or Node.js execution is required by the installed viewer.
What the plugin includes
skills/documenting-codebases/assets/documint-renderer.html— reusable original DocuMint Local UI assets, stored in Git and included in the ZIP.skills/documenting-codebases/assets/project-data.schema.json— required structured source facts.skills/documenting-codebases/scripts/render_embedded.py— optional Python standard-library JSON-to-HTML assembler. This does not run the DocuMint analyzer; a file-editing host may substitute the markers without Python.skills/documenting-codebases/references/embedded-renderer.md— safe substitution and provenance contract.
The browser interface includes project map, file finder, module architecture, dependency navigation, treemap, zoom/pan and theme controls. The UI comes from the original renderer assets, while facts extracted with an AI tool are not the same as deterministic AST analysis unless the original analyzer was separately executed. Indexed filenames are distinct from inspected file bodies.
ChatGPT trigger and format selection
Type Use DocuMint. The Skill then offers HTML, Markdown (.md) or Both as a one-time conversational choice unless the format was already specified. It uses source already available, or requests authorized source files if missing.
| Choice | Downloadable output |
|---|---|
| HTML | documentation.html — original bundled DocuMint HTML viewer with structured project JSON |
| Markdown | documentation.md — source-grounded Markdown |
| Both | documentation.html and documentation.md from one evidence model |
A skill-only plugin cannot force clickable ChatGPT checkboxes. Interactive custom forms require a supported UI extension, typically MCP-backed; this Skill intentionally needs no MCP/server and no Google Drive permissions. File downloads require file-writing tools in the host. The original HTML UI assets are reused, but AI-assisted facts and Markdown are not byte-identical to VS Code Local SourceAnalyzer outputs.
Ready-to-download HTML is mandatory
HTML selection must produce a finished `documentation.html`, not a generator script. The installed asset `skills/documenting-codebases/assets/documint-renderer.html` is the original full DocuMint viewer and is SHA-pinned as Git blob `ec66c180f09a0df8ef2eb6f3c8765b185716d8f6`. The packager rejects truncated/lookalike assets, and the renderer verifies that only project JSON and CSP nonce are substituted.
For a prior generator-only ZIP, using the installed skill (not a script inside the input archive):
```bash
python3 skills/documenting-codebases/scripts/render_embedded.py
--input-zip /path/to/DocuMint_PEVS_Original_Renderer_Package.zip
--format both --out-dir ./finished
--zip-out ./DocuMint_READY.zip
```
This extracts only `project-data.json` from that ZIP, never executes archive scripts, and writes `finished/documentation.html`, `finished/documentation.md` and a ready ZIP containing actual documentation. With no ability to access/materialize the pinned renderer or create files, report generation failure—not success with a generator package. Google Drive, MCP servers and internet connectivity are not required by the renderer.
Quick start
-
Use ChatGPT/Codex with authorized source access to prepare
project-data.jsonper the bundled schema. Make every claim evidence-based; report inspected scope, revision and unknowns. -
Select HTML/Markdown/Both. For HTML, embed JSON into the bundled original template (base64 payload plus fresh CSP nonce); Markdown is assembled from the same evidence. Where Python is available:
python3 scripts/render_embedded.py --data project-data.json --format both --out-dir ./documint -
Download only the requested file(s). If HTML was selected, open
documentation.htmlin Firefox/Chrome offline. No DocuMint executable, server, Node or CDN is needed at viewing time.
File-output limitation: A skill does not by itself grant code execution or file-writing tools. If the ChatGPT host cannot access the packaged asset or create a file, supply verified JSON/Markdown and disclose the limitation; do not invent an alternative viewer and call it DocuMint. See SKILL.md.
Maintaining the embedded UI snapshot
From the repository checkout (build time only, Node 22+):
node tools/export-embedded-renderer.mjs --check
# After canonical renderer changes:
node tools/export-embedded-renderer.mjs --write
python3 tools/documint_skill_plugin.py check
python3 tools/documint_skill_plugin.py package --output documint-embedded-skill.zip
The existing documint-cli.cjs remains optional for Codex/local workflows and direct AST-based Local Mode; it is not bundled as a Skill requirement in v0.3.2.
Privacy: PRIVACY.md · Terms: TERMS.md · Submission: SUBMISSION.md · Issues: https://github.com/Wonderer-Tech/documint/issues.