Agent Skills: Single HTML Page

Single HTML Page

Create one HTML file that can be shared without its original asset folder or live network dependencies.

Default Path

Use scripts/single_html_page.py (EXECUTE) unless the user asks to do the bundling manually:

python3 scripts/single_html_page.py input.html
python3 scripts/single_html_page.py https://example.com/page --output page-shareable.html

The script wraps monolith, removes leftover network hint and base links, and verifies that asset-bearing references are embedded as data URIs.

Tool Choice

Prefer monolith from PATH. It is the primary bundler because it embeds linked CSS, images, JavaScript, and fonts into one HTML document. If monolith is missing, check the local tool manager first (nix/home-manager or mise) instead of installing an ad hoc replacement.

For local HTML files, the script sets the base URL to the file's directory so relative assets resolve correctly. For URLs, the script lets monolith fetch the page and its assets directly.

Workflow

1. Run the script and write *-shareable.html unless the user names an output.
2. Read the script's verification summary. If it reports remaining asset refs, inspect and either fix them or tell the user what still depends on network or local files.
3. For user-facing pages, render at least one screenshot. Use desktop and mobile screenshots when responsive layout matters.
4. Preserve page content and behaviour. This skill packages the page; it does not restyle, rewrite, or redesign it.

Verification

Accept external links in normal anchors, but do not accept asset-bearing references such as img src, script src, stylesheet/icon link href, srcset, CSS url(...), or @import unless they are data: or fragment references.

If the page relies on JavaScript-rendered DOM that is not present in the source HTML, render or dump the browser DOM first, then run the bundler on that HTML with the original page URL as the base URL.

monolith -I isolates the document from the Internet, so a page that fetches at runtime (JS fetch/XHR, SPA data loads) will not work offline. Asset-ref verification cannot detect this, so rely on the step 3 screenshot to catch it.

The CSS url(...)/@import check scans the whole document, so a url(...) literal inside a <script> body or attribute string can surface as a false positive; inspect flagged references rather than trusting the exit code alone.

The bundler's pure helpers (base-URL derivation, output naming, base-tag and network-hint stripping, asset-ref detection) are pinned by tests/test_single_html_page.py - uv run --with pytest -- pytest tests/ from the skill dir.