matrize-brief
Use to synthesise a design-system derivation into the approval artefact a human signs: target taxonomy, chosen direction, what is deliberately NOT adopted and why, and falsification criteria stating what would show the system does not hold. Trigger on 'write the design brief', 'what are we actually proposing', 'put this in front of the decision-maker', 'matrize brief'. Stops outright if any discovery artefact is missing, warns when the state-of-the-art survey is past its expiry, and writes nothing further until the approval block is signed — no objection is not approval.
- License 许可
- MIT
- 最近更新
- 2026-11-02
- Stars
- ★ 1
- Forks
- 0
01 前言 Front Matter
- name
- matrize-brief
- description
- Use to synthesise a design-system derivation into the approval artefact a human signs: target taxonomy, chosen direction, what is deliberately NOT adopted and why, and falsification criteria stating what would show the system does not hold. Trigger on 'write the design brief', 'what are we actually proposing', 'put this in front of the decision-maker', 'matrize brief'. Stops outright if any discovery artefact is missing, warns when the state-of-the-art survey is past its expiry, and writes nothing further until the approval block is signed — no objection is not approval.
- argument-hint
- <system> [direction]
03 SKILL.md 正文
Synthesise discovery into a document a human can approve or reject. This is the human-in-the-loop gate, and it is the reason an n-proposal portfolio is a decision aid rather than a procrastination machine.
Hard input gate
Read <root>/system/DECODE.md and <root>/system/LEXIKON.md first. If either is
missing, say so and stop. They come from matrize-decode and matrize-name; run
those. Do not synthesise around an absent input — a brief assembled from a missing
measurement is the one artefact nobody can audit later.
If <root>/system/STATE-OF-THE-ART.md exists and is past its stated expiry, warn and
name the claim that forced it. The style of a design system does not go stale; its
delivery chain does.
What the brief contains
- Target taxonomy — the named elements this system will carry, from
LEXIKON.md. - Chosen direction, stated in one paragraph a non-technical reader can hold.
- What is deliberately not adopted, and why. This section is not optional and not decorative. A brief that only says yes has recorded no decision.
- Falsification criteria — what would show this system does not hold. Write these as things that could actually be observed: "a view needs two dominant action colours and neither can be demoted", "the spacing scale cannot express the densest table we ship". A criterion nobody could ever observe is not a criterion.
- Open questions, carried verbatim from every grade-C-only finding and every
unanswered
preflightquestion. - Coverage — state explicitly which of the six required areas (states, contrast,
motion, inverse mode, content rules, provenance) the system covers and which it does
not. A system missing any of them is incomplete, and this is where that is said.
Check the asset classes too, against
${CLAUDE_PLUGIN_ROOT}/references/vocabulary/visual-asset-taxonomy.md— three are missing from almost every design system and almost always needed: empty state, error state, and Open Graph image. Treat them as mandatory coverage.
Phase entry criteria are gates, not aspirations
The three build methods — echo, spread, retrofit — read this brief and treat its
entry criteria as binding. So write them as checkable preconditions: "tokens.json
validates against the DTCG profile", "every colour role has a computed contrast figure",
not "colours are finalised".
Tell the approver plainly that they steer execution by editing this file. An edited entry criterion is honoured; a note in a chat is not.
The approval block
End with it, and enter plan mode if the session supports it:
This is a discussion basis, not a finished rulebook.
Decision-maker: ______________________
Gate criteria: feedback on <the specific things> before production
Approved by: ________________ Date: __________
Approval covers: <direction only> | <direction and taxonomy> | <the whole system>
Then stop — write nothing further until the user explicitly approves. "No objection"
is not approval. matrize-status reads this block and reports whether it is signed.
This brief is a hypothesis
Say so in the document. What the first real emit surfaces — a token the taxonomy cannot express, a role that turns out to be two roles, a reference that contradicts another — is expected to revise this brief. A regenerated brief after the first build is the normal path, not a correction.
Staleness
Compare modification times. If any input is newer than an existing BRIEF.md, the brief
is being justifiably regenerated. If the brief is newer than every input and it is being
re-run anyway, ask what changed. Either way, record the input timestamps in the header so
a reviewer can see what it was built from.
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.