AGENT SKILL

brand-audit

Reverse-engineer an imprint's target design image into the imprint's references/visual-audit.md (palette with sampled hex and approximate ratios, typography roles, spacing, composition, navigation, image behavior, mark, line weights, texture, hierarchy, repeated motifs), classifying every observation as core identity, candidate motif, or hero-specific, each a numbered proposal for the owner to check at PR review. Also freezes the target image from Notion into references/target-homepage.png. Trigger on "/brand-audit", "audit the Valesor target", "what's intentional in the S9 homepage", "freeze the new target image", "re-audit after the new mockup". Requires the visual-designer role. Not for writing tokens or brand docs from an approved audit (visual-designer, directly), the illustration system (art-direction), checking built pages (brand-qa), or a game's concept images (that game's concept-art role).

License 许可
Apache-2.0
最近更新
2026-10-28
Stars
★ 0
Forks
0

01 前言 Front Matter

name
brand-audit
description
Reverse-engineer an imprint's target design image into the imprint's references/visual-audit.md (palette with sampled hex and approximate ratios, typography roles, spacing, composition, navigation, image behavior, mark, line weights, texture, hierarchy, repeated motifs), classifying every observation as core identity, candidate motif, or hero-specific, each a numbered proposal for the owner to check at PR review. Also freezes the target image from Notion into references/target-homepage.png. Trigger on "/brand-audit", "audit the Valesor target", "what's intentional in the S9 homepage", "freeze the new target image", "re-audit after the new mockup". Requires the visual-designer role. Not for writing tokens or brand docs from an approved audit (visual-designer, directly), the illustration system (art-direction), checking built pages (brand-qa), or a game's concept images (that game's concept-art role).
allowed-tools
Bash(.claude/bin/role:*) Bash(git log:*) Bash(git fetch:*) Bash(python3 -c:*) Bash(magick:*) Bash(identify:*)
argument-hint
<slug> [freeze <notion-url> | revise]

02 资源文件 2

03 SKILL.md 正文

Brand audit

Request: $ARGUMENTS

The target is evidence, not a spec. This audit turns what it shows into decisions the owner makes, one observation at a time. Nothing downstream (tokens, brand docs, components) may cite the target directly: it cites the approved audit.

Preflight

  1. Run .claude/bin/role require visual-designer ${CLAUDE_SESSION_ID}. If it fails, stop and tell the owner to run /role visual-designer.
  2. git fetch origin. Resolve the slug (brands/<slug>/). Read the imprint's Notion "Brand Kit Project Direction" page (the role file or brand/README.md links it), any existing brand/ docs, and a previous visual-audit.md.

Freeze the target (freeze, or when references/target-homepage.png is missing)

  1. Fetch the image from the Notion page (the file block's download URL), and save it as references/target-homepage.png. Never re-encode or resize it.
  2. Record its provenance in references/README.md: the Notion page URL, file name, pixel size, date fetched, and sha256.
  3. A target that changes later is a new freeze in its own commit, and the audit gets a revision against it. Never overwrite silently.

Audit

Open the image and look at it, at full size and zoomed into each region. Never describe it from the Notion text alone. Fill references/audit-template.md:

  1. Palette: sample every distinct colour (python3 -c with Pillow, or magick … -unique-colors, on regions you name), with hex values and approximate area ratios. Note anything that looks like a generator artefact (off-by-a-few greys, noise) rather than a colour decision.
  2. Typography: each text role (wordmark, display, body, mono metadata, links, nav), with a likely typeface family or class, size ratio to body, weight, case, and tracking. Name candidates for open-source, self-hostable fonts, but choosing one is the owner's call.
  3. Layout: grid and columns, max width, gutters, vertical rhythm, rule weights and where rules carry the layout.
  4. Composition and image behavior: how the hero sits (bleed, mask, empty zone for type), crop behavior implied for mobile.
  5. Navigation, mark, components: header, nav, CTAs, cards and rows, metadata blocks, footer. Each repeated pattern is a component candidate.
  6. Line, texture, hierarchy, motifs: line weights, texture layers, the read order, and anything repeated.
  7. Copy: headlines, CTA wording, and voice signals visible in the image.
  8. Tells: pseudo-text, broken letterforms, impossible geometry, misspelt names, wrong URLs. These are never core.

Classify

Every observation gets one class, marked [PROPOSED P<n>] (the next IDs unused in brands/<slug>/, per the role's Proposals section), with one line of reasoning:

  • core identity: persists across every application of the brand
  • candidate motif: worth testing in other applications before adopting
  • hero-specific: belongs to this composition only (it goes to a slot brief, never to the brand docs)

Test each proposed core item against the Notion page's direction (its "character", "not", and anti-pattern lists) and against the sibling imprints' approved audits. A core item that matches a sibling's core is a distinctness flag: say so.

Ship

The audit goes in its own PR on a design/ branch, labeled role:visual-designer, after /pre-pr. The PR body has one checkbox per proposal, core items first, each at the start of a line: - [ ] P<n>: <class> — <observation> (references/visual-audit.md#<section>). The owner decides at PR review: a checked box approves that ID, and one still unchecked at merge stays [PROPOSED P<n>]. Never tell them the audit is approved when merged. Record a reclassification or change they ask for in the audit's Decisions section, on the same branch. Checked IDs become [DECIDED <YYYY-MM-DD>, PR #<pr>, P<n>] after the merge (the role's Proposals section), never in the open PR.

Never write tokens/, brand/visual-language.md, or component specs from an unmerged audit. Stop after the PR.

Output

The PR link, the count of core / candidate / hero-specific items, the distinctness flags, open questions for the owner (at most five, each with a recommendation), and one next action, usually "review the audit PR", or after merge, "derive tokens from the checked core palette and type".

04 同仓库更多 Skills

valesordev
valesordev/automate.valesordev.com

up-skill

Upstream skill, for when the dependency's job is asked for. Not for local work.

★ 0
valesordev
valesordev/automate.valesordev.com

do-thing

Do the thing when asked to do the thing. Not for other things.

★ 0
valesordev
valesordev/automate.valesordev.com

sre-start-sprint

Work the active sprint as SRE, with stories as GitHub issues. Observability review of the sprint's drafts first (architecture's contract review waits on it), then instrumentation verification for stories at review, then SRE's own backlog. Run once PM has activated the sprint; it also resumes a sprint already under way.

★ 0
valesordev
valesordev/automate.valesordev.com

sprint-run

Plan the active sprint's role lanes and, once the owner confirms, start them in parallel. Reads the board, shows which stories and queues each role the repo runs as a lane (implementation, architecture, SRE, PM, and where the charters say so visual-designer and product) can pick up now and what everything else is waiting on, then runs one isolated `claude -p` session per lane (own worktree, role set, hooks on) via `.claude/bin/coordinator`, and reports from live state. Trigger on "run the sprint", "/sprint-run", "start the lanes", "what can the agents pick up", "kick off implementation and SRE". Not for working one story yourself (impl-start-sprint), activating or closing a sprint (pm-start-sprint, pm-close-sprint), or merging.

★ 0
valesordev
valesordev/automate.valesordev.com

role

Set this session's role (e.g. pm, product, architecture, sre, implementation, visual-designer, or none) and load its charter, then start an optional task given after the role; or show the current role when run with no argument, or the repo's role catalog with `help`. The owner runs it at session start. Never invoke it on your own or to change roles mid-task.

★ 0
valesordev
valesordev/automate.valesordev.com

release-review

Review a release's readiness across every delivery planner's board and repo (each feature's success signal, the milestones and stories delivering it, blockers and who owns them) and recommend go, no-go, or cut, written to docs/product/reviews/ (one file per review). Trigger on "/release-review", "where are we on R1", "are we ready to ship", "what's blocking the release", "go/no-go", and when a planner closes the last milestone a release names. Requires the product role. Not for sprint or cycle status (project-sync, producer-close-cycle), changing release scope (release-plan), or shipping or announcing anything (the owner).

★ 0