matrize-emit
Use to render a derived design system to a concrete target by running the plugin's committed formatter scripts against its DTCG token file — html and pdf produce the landscape sketchbook that carries the approval block, css and vitepress produce platform output, and provenance produces the graph viewer. Trigger on 'emit the design system as CSS', 'produce the sketchbook PDF', 'regenerate the docs theme from tokens', 'matrize emit'. One neutral source, N formatters: nothing here is authored per run, so the same tokens regenerate byte-stable, and prose from an R2 or R3 reference is refused rather than reproduced.
- License 许可
- MIT
- 最近更新
- 2026-11-02
- Stars
- ★ 1
- Forks
- 0
01 前言 Front Matter
- name
- matrize-emit
- description
- Use to render a derived design system to a concrete target by running the plugin's committed formatter scripts against its DTCG token file — html and pdf produce the landscape sketchbook that carries the approval block, css and vitepress produce platform output, and provenance produces the graph viewer. Trigger on 'emit the design system as CSS', 'produce the sketchbook PDF', 'regenerate the docs theme from tokens', 'matrize emit'. One neutral source, N formatters: nothing here is authored per run, so the same tokens regenerate byte-stable, and prose from an R2 or R3 reference is refused rather than reproduced.
- argument-hint
- --target <html|pdf|css|vitepress|provenance>
03 SKILL.md 正文
Run a formatter. Do not author one at run time.
emit is a tool-wrapper: the formatters are committed scripts under
${CLAUDE_PLUGIN_ROOT}/scripts/, so emitting is deterministic and free. Authoring a new
formatter is a change to this plugin, reviewed like any other — not something that
happens invisibly inside a run and produces slightly different output each time.
That determinism is what makes matrize-retrofit's zero-visual-diff proof possible at
all. A formatter that re-decides its output per run cannot prove equivalence with
anything.
Input gate
Requires <root>/system/tokens.json. If it is absent, say which phase produces it
(decode + name, or retrofit) and stop.
If a spread choice record exists and is unanswered, the PreToolUse guard denies the
write — so record the choice first. That is a refusal, not a warning, and reaching for
MATRIZE_DISABLE_GUARD=1 to get past it defeats the phase.
The rights refusal
Reference prose from an R2 or R3 source is never written into an emitted artefact.
Values, measured facts, and rules restated in this system's own words are fine. The source's sentences, illustrations and assets are not. When a formatter would need to reproduce reference text to be useful, the design is wrong — cite and link instead.
Targets
html is the reference formatter and simultaneously the portfolio: the landscape
sketchbook, one building block per page, marginal callouts carrying purpose, rule and
anti-rule, and the approval block naming the decision-maker and the gate criteria.
pdf renders that HTML through headless Chrome --print-to-pdf, honouring
@page { size: … landscape }.
Never wkhtmltopdf. Its Qt WebKit engine renders @font-face spans blank — text
disappears entirely while spaces are preserved — does not interleave multi-page CSS Grid
columns, silently ignores header/footer flags on unpatched builds, and has no useful
--zoom when the CSS uses physical units. Those are documented failures, not
speculation, and re-acquiring them would be a choice.
Verify the render, do not trust it. After producing a PDF: extract its text
(pdftotext -layout) and grep for strings that must be present. A silently blanked span
looks fine in a page count and is invisible in a thumbnail. Also confirm the page count
and reading order.
The targets that exist, and the ones that do not
A target is real when a committed script emits it. Anything else is a plan, and naming it here as though it worked would be the same defect as a rule with no card behind it.
| target | script | status |
|---|---|---|
css |
scripts/emit_css.py |
ships |
vitepress |
scripts/emit_vitepress.py |
ships |
html |
scripts/build_sketchbook_html.py |
ships |
pdf |
the html output through headless Chromium |
ships |
provenance |
scripts/build_provenance_html.py |
ships |
tailwind, scss, json, toml |
— | not built. Do not offer them |
If a user asks for one of the last row, say it is not implemented and name what is. A formatter is a small, well-specified piece of work; pretending one exists is not.
css and vitepress are platform output. Neither is ever an input: if a change needs
making, it is made in tokens.json and re-emitted. A hand edit to emitted output is lost
on the next run, and worse, it makes the source untrue.
Report
Name the target, the formatter script that ran, the output path, and — for pdf — the
verification that actually executed, with its result. If a verification step did not
run, say so rather than implying it passed.
04 同仓库更多 Skills
toy-check
A skill that checks things in the repository and reports what it finds.
demo-check
Reports which generated files changed since the last generation run. Use when the user asks what drifted. Read-only; never edits a file.
demo-check
Reports which generated files changed since the last generation run. Use when the user asks what drifted. Read-only; never edits a file.
demo-check
Reports which generated files changed since the last generation run. Use when the user asks what drifted. Read-only; never edits a file.
demo-check
Reports which generated files changed since the last generation run. Use when the user asks what drifted. Read-only; never edits a file.
zirkel-verify-assumptions
Checks exactly ONE named assumption against real evidence in at most 3 steps and reports one outcome: confidence raised (>=90), reading changed, or still unresolved. Use to resolve a single load-bearing uncertainty — typically a blocking entry surfaced by zirkel-clarify-scope: "verify this one assumption", "is it actually true that X", "check this before we rely on it". For multiple uncertainties, invoke once per uncertainty — the step budget is never shared.