The CLI contract

This page is the SPEC, rendered by the tool the SPEC specifies. Transcribed from SPEC.md at docs-build time — the spec documents itself.

CLI contract

dogbed render <template.knap> [--data <file>] [--profile html|xhtml] [--title <text>] [--css <href>] [--max-output <bytes>]
--data - reads the JSON object from stdin: cat findings.json | dogbed render verdict.knap -d - > verdict.html
Output is a bare HTML fragment unless --title or --css is passed. Either flag wraps the fragment in a minimal shell: HTML5 normally, XHTML 1.0 Strict under --profile xhtml. The title is HTML-escaped; the css hrefs pass through verbatim — URL or relative path, the consumer problem, not ours. No flags means byte-identical fragment output.
--max-output caps the final document in bytes, shell included (it also bounds the k4o stage; default 256 MiB, 0 = unlimited). Nested loops multiply, so the default is generous but finite.
dogbed template <name> prints an embedded starter template to stdout so the user copies it and owns it. The library is exactly three starters — verdict, release-notes, reading-note — with matching example data. No discovery paths, no versioning, no fetching.
Errors go to stderr, stdout stays empty, exit 1. A failed render never emits a half-rendered document.

Markdown input to the pipeline (oliver already does that alone; dogbed is about templated documents).
HTML4 Strict output (declined upstream; the bar is a named consumer).
Any CMS feature, ever.
Template versioning, remote templates, template discovery paths — the starter library is three embedded templates, done.

It is a compiler: deterministic, pipeable, boring in the good way. If it ever grows an edit button, kill it.