Design Brief — Orchestrator

Communication — visual layer. Produces a graphic-design brief for a single visual asset; rendering is downstream.

Core Question: "Could a designer or image-gen tool execute this asset on-brand and on-platform without follow-up questions?"

Critical Gates — Read First

• Do NOT render. This skill produces the brief, not the asset. The brief carries spec + reference direction + image-gen prompt (where applicable). Rendering happens downstream — image-gen (Midjourney / Imagen / DALL·E / Claude Design), vector tooling (Pencil / Figma), or a human designer.
• Do NOT proceed without brand anchors. Missing brand/BRAND.md or brand/DESIGN.md → return NEEDS_CONTEXT, recommend running brand-system first.
• Do NOT invent tokens, fonts, or motion specs. Every visual decision traces to DESIGN.md. If DESIGN.md doesn't cover what's needed (e.g. illustration style), flag it in the brief — don't guess.
• Do NOT use stock-AI defaults. No default purple-blue gradients, centered-isolated-on-white, faux-3D bevels, or glassmorphism unless DESIGN.md specifies. Critic scores generic-AI smell explicitly.
• Do NOT skip the brief approval gate. The brief is a candidate, not a delivery — user reviews before downstream rendering.
• Platform spec is mandatory. Every brief includes aspect ratio, safe zones, mobile readability (type scale, thumb-stop contrast), file format, and file-size limits. See references/platform-modules.md.

Philosophy

A great brief eliminates downstream ambiguity. Render quality is bounded by brief quality — a vague brief produces drift no matter how good the renderer. The brief is the deliverable; every visual decision binds to a brand anchor and a platform constraint.

Brand fidelity > aesthetic novelty. Platform fitness > generic polish. A boring on-brand brief beats a striking off-brand one.

Inputs Required

• Asset request — type (IG carousel, OG image, banner ad), platform/format, purpose (announce/educate/convert/recruit), copy if available
• brand/BRAND.md — voice, archetype, sacred elements
• brand/DESIGN.md — palette, typography, surface language, motion (if applicable)

Inputs Optional

• brand/ASSETS.md — pre-fills format/dimensions if asset matches a row; ticks box on completion
• .agents/skill-artifacts/mkt/lp-brief/[slug]/asset-slots/[slot-id].md — slot spec when brief is for a landing-page asset
• .agents/skill-artifacts/mkt/content/[slug].copy.md — copy used IN the asset (headline, body, CTA)
• .agents/skill-artifacts/mkt/campaign-plan.md — campaign context, channel placement, awareness stage

Output

.agents/skill-artifacts/mkt/design-briefs/[slug].md containing:

• Approved concept (visual direction, references)
• Brand anchors (palette/typography/sacred elements from DESIGN.md)
• Platform spec (aspect ratio, safe zones, type scale, contrast, format, size cap)
• Asset slots (compound assets — e.g. carousel slides)
• Copy placement (when copy lives in the asset)
• Failure modes (platform-specific + generic-AI)
• Downstream handoff: image-gen prompt OR designer spec OR vector-tool spec
• Critic report (rubric + generic-AI-aesthetic check)

Quality Gate

Before delivering, the critic agent verifies:

• Brand fidelity — palette, typography, motion all trace to DESIGN.md
• Sacred elements respected (no proposed change to logo, primary palette anchor, tagline, etc. unless brief explicitly authorizes)
• Hierarchy — clear focal point, scannable in 1 second
• Composition — balance, intentional white space
• Typography — pairing/sizing/leading consistent with DESIGN.md
• Contrast — text passes WCAG AA (≥4.5:1 normal, ≥3:1 large) on its actual background
• Platform fit — dimensions, safe zones, platform crop behavior, mobile thumb-stop readability, file format, file-size cap all verified against the asset's platform module
• CTA clarity (if applicable) — readable at preview size, action verb visible
• Generic-AI-aesthetic check — full 13-pattern detector from references/visual-rubric.md §"Generic-AI-Aesthetic Detector". Score 0-3 per pattern (max 39). Thresholds: 0-7 clean / 8-15 DONE_WITH_CONCERNS / 16+ auto-FAIL.
• Downstream-handoff completeness — image-gen prompt is specific (lens / lighting / mood / era / composition / color cast), or designer spec covers placement/typography/color tokens, or vector-tool spec specifies layout grid

Chain Position

Previous: brand-system (required), lp-brief (optional — slot spec), copywriting (optional — copy) | Next: external rendering (image-gen / Pencil / Figma / human designer)

Re-run triggers: BRAND.md or DESIGN.md updates, new asset row in ASSETS.md, lp-brief slot needs an explicit per-asset brief, campaign launch.

Skill Deference

• No brand system? → run brand-system first; design-brief hard-blocks without it.
• Need in-asset copy? → run copywriting first or in parallel.
• Whole-page redesign? → use lp-brief.

---

Agent Manifest

Agent	Layer	File	Focus
Brand-Anchor Agent	1 (parallel)	agents/brand-anchor-agent.md	Pulls relevant tokens, sacred elements, lexicon, motion spec from BRAND.md + DESIGN.md
Concept Agent	1 (parallel)	agents/concept-agent.md	Generates 3 distinct concept directions (mood, composition, references)
Copy-Anchor Agent	1 (parallel)	agents/copy-anchor-agent.md	Resolves copy that appears IN the asset (from copywriting artifact, or interview the user)
Brief Synthesizer	1.5 (after L1)	agents/brief-synth-agent.md	Merges anchor + 3 concepts + copy into 3 candidate briefs with platform spec, hierarchy, asset slots
Prompt-Craft Agent	2 (downstream-route: image-gen / template-pack)	agents/prompt-craft-agent.md	Produces image-gen prompts (Claude Design / Midjourney / Imagen / DALL·E / Veo / Suno) for the chosen brief
Figma-Spec Agent	2 (downstream-route: designer-handoff)	agents/figma-spec-agent.md	Produces design spec markdown for human designer in Figma
Critic Agent	3 (final)	agents/critic-agent.md	Visual rubric scoring + generic-AI-aesthetic detection + platform-fit check

---

Routing Logic

Every brief carries a downstream-route tag identifying its renderer. The tag drives which optional sub-agent runs after the brief is approved.

Downstream routes

Route	When	Optional Layer 2 agent	Output addition
image-gen	Photographic, illustrative, abstract, or compositionally complex. Hero/OG/blog illustrations, ad backgrounds, video thumbnails.	prompt-craft-agent	Image-gen prompt + 2 variants
vector-tool	Vector layouts, UI mockups, branded social templates, multi-format variants, infographics. Pencil/Figma execution.	(none — brief carries vector-tool spec block)	Layout grid spec + token references
designer-handoff	Executed by human designer in Figma or print shop. Print-grade, OOH, complex composition.	figma-spec-agent	Designer-file spec
template-pack	Multi-format social packs (IG + LinkedIn + X variants of one asset).	prompt-craft-agent per format	Per-format prompts in one brief

Override auto-detection with --route=image-gen|vector-tool|designer-handoff|template-pack.

Auto-detection (asset type → default downstream route)

Asset Type	Default Route	Why
OG image / blog hero / ad photo	image-gen	Generative gives photo/illustration diversity
Instagram carousel (typographic)	vector-tool	Vector + multi-slide layout
Instagram carousel (image-led)	image-gen (per slide)	Generate slides, assemble
Instagram post / story	image-gen or vector-tool	Depends on photo vs typographic
LinkedIn document post	vector-tool	Multi-slide typographic
LinkedIn single-image	image-gen or vector-tool	Depends on photo vs typographic
FB / display banner (text-heavy)	vector-tool	Vector text + crop variants
FB / display banner (visual-led)	image-gen	Generate visual, overlay text
YouTube thumbnail	image-gen	Photo-based + bold text overlay
X/Twitter card	image-gen or vector-tool	Depends on photo vs typographic
OOH / billboard / print	designer-handoff	Print-grade, needs human designer
Email hero	image-gen or vector-tool	Depends on photo vs typographic
Hero illustration (custom)	image-gen or designer-handoff	Generative or human-designer
Spot icon / decoration	vector-tool	Vector

---

Pre-Dispatch

This skill is hard-gated on brand artifacts. Cold-start questioning happens after the gate. Full Pre-Dispatch protocol pattern: meta-skills/references/pre-dispatch-protocol.md.

Hard gate (before any questioning)

brand/BRAND.md AND brand/DESIGN.md must be present. If either missing → return NEEDS_CONTEXT, recommend brand-system (Route A for BRAND only, Route B for full design tokens). If either >60 days stale and user hasn't confirmed, warn before proceeding.

If brand/ASSETS.md exists, scan for a row matching the requested asset. If found, pre-fill format/dimensions/path and prepare to tick the checkbox on completion.

Needed dimensions

• Asset type (OG image / IG carousel / banner / hero / OOH / etc.)
• Downstream route (image-gen / vector-tool / designer-handoff / template-pack — drives Layer 2)
• Brand reference (resolved by hard gate to brand/DESIGN.md)
• Copy/headline if any (to render IN the asset)
• Constraints (dimensions, deadline, must-include elements)

Read order (post-gate)

1. Pipeline: brand/BRAND.md, brand/DESIGN.md (confirmed by hard gate). brand/ASSETS.md for dimension pre-fill. .agents/skill-artifacts/mkt/lp-brief/[slug]/asset-slots/[slot-id].md if invoked from lp-brief. .agents/skill-artifacts/mkt/content/[slug].copy.md if copy supplied separately. .agents/skill-artifacts/mkt/campaign-plan.md for campaign context. research/icp-research.md for audience visual preferences.
2. Experience: .agents/experience/{brand,goals}.md.

Optional Artifacts (read if present)

Artifact	Source	Benefit
brand/ASSETS.md	brand-system Route B	Auto-fill dimensions, tick checkbox on completion
.agents/skill-artifacts/mkt/lp-brief/[slug]/asset-slots/[slot-id].md	lp-brief	Slot spec when brief is for an LP asset
.agents/skill-artifacts/mkt/content/[slug].copy.md	copywriting	Copy to use in the asset
.agents/skill-artifacts/mkt/campaign-plan.md	campaign-plan	Campaign context, awareness stage
research/icp-research.md	icp-research	Audience visual preferences

Warm Start (invoked from lp-brief or campaign-plan with asset spec):

Hard gate passed: brand/BRAND.md + brand/DESIGN.md present.
Found:
- asset spec → "[from upstream slot or campaign-plan]"
- copy → "[from copywriting if present]"
- ASSETS.md row → "[matched / no match]"

Auto-detecting downstream route from asset type. Override or proceed?

Cold Start (standalone invocation, asset request from user):

design-brief produces a per-asset brief that downstream renderers (Claude
Design / Pencil MCP / Figma / human designer) execute. Before I dispatch:

1. **Asset type** — what's being designed? (OG image / IG carousel / IG post /
   IG story / LinkedIn document / LinkedIn single-image / FB ad / YouTube
   thumbnail / X card / OOH / banner / hero / other.)
2. **Downstream route** — image-gen (generative for photos/illustrations) /
   vector-tool (Pencil/Figma for layouts and variants) / designer-handoff
   (human designer for print or complex composition) / template-pack
   (multi-format social packs from one brief). Auto-detected from asset
   type by default — override here.
3. **Copy/headline** — what text appears IN the asset? Headline, body,
   CTA, brand mark text. Reference `.agents/skill-artifacts/mkt/content/[slug].copy.md`
   if supplied separately.
4. **Constraints** — dimensions (if non-standard), deadline, must-include
   elements (logo placement, brand mark, legal disclaimer, etc.).

Answer 1-4 in one response. (Brand reference is auto-resolved from
brand/BRAND.md + brand/DESIGN.md.) I'll dispatch.

Write-back:

Q	File	Key
4. Constraints (durable: legal disclaimer policies, logo placement rules)	brand.md	Brand — design constraints (only if user expresses durable rule)
1, 2, 3. Asset-specific dimensions	(per-asset, lives in design-brief artifact)

---

Step 0.5: Route Detection (Orchestrator)

Before Layer 1, pick a downstream route AND (for image-gen) a target generative tool. Both are required by downstream sub-agents.

1. --route= override → honor it. Warn if asset-type default differs.
2. Else walk the asset-type → default route table above.
3. State route + rationale in 1 line before dispatching:

   "Downstream route: image-gen (photographic OG image). Target tool: midjourney-v6 (editorial mood). Override with --route= if needed."

4. For image-gen, pick target_tool from references/prompt-patterns.md "tool → asset type" table (e.g., claude-design, midjourney-v6, imagen-3, dall-e-3, ideogram, veo, suno).
5. Pull the platform module from references/platform-modules.md and pass its spec checklist (aspect, safe zones, type scale, contrast, format, size cap, anti-patterns) to brief-synth-agent and critic-agent.

route, target_tool, and platform_module are part of every agent's context.

---

Layer 1: Parallel Foundation

Spawn IN PARALLEL; outputs feed Layer 1.5:

Agent	Instruction File	Pass These Inputs	Reference Files
Brand-Anchor Agent	agents/brand-anchor-agent.md	full BRAND.md + DESIGN.md + asset request	—
Concept Agent	agents/concept-agent.md	asset request + brand digest	references/asset-types.md, references/platform-modules.md, references/failure-modes.md
Copy-Anchor Agent	agents/copy-anchor-agent.md	asset request + copywriting artifact (if any) + brand voice rules	—

---

Layer 1.5: Brief Synthesis

Agent	Instruction File	Pass These Inputs	Reference Files
Brief Synthesizer	agents/brief-synth-agent.md	brand anchor + 3 concepts + copy + route + platform module + asset request	references/asset-types.md, references/platform-modules.md

Output: 3 candidate briefs, each with concept name, visual direction, hierarchy, platform spec (resolved against the module), copy placement, failure modes to avoid.

---

Approval Gate 1 — Brief Selection

STOP and present the 3 candidate briefs. Do not proceed to image-gen prompt or designer spec.

Format:

## Brief Candidates

### A. [Concept Name]
[Visual direction, mood, references — 3-5 lines]
**Why this:** [argument]
**Risk:** [main concern]

### B. [Concept Name]
...

### C. [Concept Name]
...

**Pick one (A/B/C), request revisions, or specify your own direction.**

User responses:

• "A" / "B" / "C" → proceed with that brief to Layer 2.
• "Revise X" → re-dispatch concept-agent with feedback, regenerate, re-present.
• "None of these" → ask one clarifying question, regenerate.
• "Switch route to X" → re-dispatch brief-synth with the new route. If concept-agent's tool-feasibility for that concept was RETHINK on the new route, re-run concept-agent first.
• "Stop" → save as .agents/skill-artifacts/mkt/design-briefs/[slug]-candidates.md, exit BLOCKED.

---

Layer 2: Downstream Handoff Augmentation

Dispatch ONE based on downstream route:

Route	Agent	Output addition
image-gen	agents/prompt-craft-agent.md	Image-gen prompt block (primary + 2 variants) appended to brief
vector-tool	(none — brief synth already wrote the vector spec block)	—
designer-handoff	agents/figma-spec-agent.md	Designer-handoff spec block appended to brief
template-pack	agents/prompt-craft-agent.md per format	Per-format prompt blocks

---

Layer 3: Critic Gate

Agent	Instruction File	Receives
Critic Agent	agents/critic-agent.md	Approved brief + Layer 2 output + platform module

Critic produces an inline rubric report with PASS / FAIL.

• PASS → Approval Gate 2.
• FAIL → re-dispatch the relevant Layer 2 agent or brief-synth with feedback. Max 2 rewrite cycles; after 2 failures, deliver with critic annotations as DONE_WITH_CONCERNS.

---

Approval Gate 2 — Brief Acceptance

STOP and present the assembled brief + critic report.

Format:

## [Asset Name] — Design Brief — [Downstream route]

**Concept:** [name selected at Gate 1]
**Critic:** [PASS / DONE_WITH_CONCERNS]
**Score:** [/40 with sub-scores]

[Brief preview — concept, anchors, platform spec, copy placement, failure modes, handoff block]

[Critic notes — concerns to monitor]

**Approve to write artifact, revise, or reject.**

User responses:

• "Approve" →
  1. Write .agents/skill-artifacts/mkt/design-briefs/[slug].md.
  2. ASSETS.md auto-tick: if the brief's asset path is a literal string match for a brand/ASSETS.md row's path field (never auto-tick on slug or asset-type heuristic), flip [ ] → [x] and append a date stamp. No match → skip; design-brief doesn't own ASSETS.md row creation (that's brand-system).
  3. Status DONE.
• "Revise X" → re-dispatch brief-synth or Layer 2 agent with feedback (1 cycle), re-present.
• "Reject" → save as .agents/skill-artifacts/mkt/design-briefs/[slug]-rejected.md with a Rejection Notes block, exit BLOCKED.

---

Artifact Template — .agents/skill-artifacts/mkt/design-briefs/[slug].md

---
skill: design-brief
version: 1
date: [today]
status: [done | done_with_concerns | blocked | needs_context]
downstream_route: [image-gen | vector-tool | designer-handoff | template-pack]
target_tool: [claude-design | midjourney-v6 | imagen-3 | dall-e-3 | ideogram | pencil | figma | print | ...]
asset_type: [og-image | ig-carousel | ig-post | ig-story | li-doc | li-single | fb-ad | yt-thumbnail | x-card | ooh | banner | ...]
platform: [instagram | linkedin | facebook | youtube | x | print | web | email | ...]
dimensions: [WxH or per-format list]
brand_anchors:
  primary_color: [hex]
  primary_type: [font, weight]
  motion: [duration token if applicable]
sacred_respected: [list of sacred elements honored]
---

# Design Brief: [Asset Name]

**Asset:** [type + platform]
**Purpose:** [announce | educate | convert | recruit | brand-build]
**Source copy:** [path or "none"]
**Downstream route:** [image-gen / vector-tool / designer-handoff / template-pack]

## Concept (Approved)

**Name:** [concept name]
**Visual direction:** [3-5 lines — mood, composition, palette emphasis, type role, motion if applicable]
**References:** [3 reference URLs or named brands/artworks IF applicable — never copy, only direction]

## Brand Anchors

- **Palette pull:** [3-5 hex values from DESIGN.md, with token name]
- **Typography:** [primary + secondary from DESIGN.md, sizes for this asset]
- **Sacred elements respected:** [list — what was preserved]
- **Lexicon:** [forbidden phrases avoided, preferred phrases used if copy is in asset]

## Platform Spec

| Field | Value | Source (platform module) |
|-------|-------|--------------------------|
| Aspect ratio | [X:Y] | [platform module name] |
| Dimensions | [WxH px] | [platform module] |
| Safe zone | [px from edges, platform-specific] | [platform module] |
| Type scale (mobile readability floor) | [min px at 1x] | [platform module] |
| Contrast (thumb-stop) | [WCAG ratio min for the asset's smallest text on its actual background] | platform module |
| File format | [PNG / SVG / WebP / MP4 / ...] | [platform module] |
| File-size cap | [KB/MB] | [platform module] |
| Color mode | [sRGB / DCI-P3 / CMYK] | [platform module] |
| Anti-patterns flagged | [list specific to this platform] | [platform module] |

## Hierarchy

1. **Focal point:** [what the eye lands on first]
2. **Supporting:** [secondary element]
3. **Tertiary:** [text body, fine detail]

## Asset Slots (compound assets only — e.g., carousel)

| Slot | Dimensions | Format | Fallback |
|------|-----------|--------|----------|
| [Slide 1] | [WxH] | [PNG] | [...] |
| [Slide 2] | [WxH] | [PNG] | [...] |

## Copy Placement (if any)

- **Headline:** "[exact copy]" — [position, type token]
- **Body:** "[exact copy]" — [position, type token]
- **CTA:** "[exact copy]" — [position, type token, contrast pair]

## Failure Modes to Avoid

- [Platform-specific traps — from platform module]
- [Generic-AI smell — from failure-modes.md]
- [Brand drift — from brand_anchors]

## What NOT to Do

- [Sacred elements: do not propose changing X, Y, Z]
- [Off-brand defaults to reject]

## Downstream Handoff Block

### When `downstream_route: image-gen`
- **Primary prompt** ([target_tool]): [full prompt with lens / lighting / mood / era / composition / color cast / aspect-ratio flag]
- **Variant 1:** [specific deviation]
- **Variant 2:** [specific deviation]
- **Post-processing note:** [overlay copy, logo placement, export profile]

### When `downstream_route: vector-tool`
- **Layout grid:** [columns × rows, gutters, baseline]
- **Component references:** [DESIGN.md tokens used]
- **Multi-format crops** (if applicable): [per-format dimensions and reflow rules]

### When `downstream_route: designer-handoff`
- **Spec sheet** for the designer (Figma file or print shop): [section list, type scale, palette, asset bundle paths]
- **Open questions for designer:** [decisions deferred to designer judgment]

### When `downstream_route: template-pack`
- **Per-format prompt blocks:** [one per platform variant]

## Critic Report

[rubric scores, PASS / FAIL, concerns to monitor]

On re-run with same slug: rename existing artifact to [slug].v[N].md and create new with incremented version. Preserves history for A/B comparison.

---

Anti-Patterns

Skipping the brief approval gate — Going straight to image-gen burns credits or designer time on a misaligned concept. INSTEAD: 3 briefs, user picks one, then handoff.

Inventing tokens — Using a color or font not in DESIGN.md because it "fits the mood." INSTEAD: If DESIGN.md is incomplete for this asset's needs, flag it in the brief and ask the user — either expand DESIGN.md (re-run brand-system) or accept the limitation.

Stock-AI defaults — Default purple-blue gradient, centered isolated subject on white, glassmorphism, faux-3D bevels. INSTEAD: Critic-agent's generic-AI rubric explicitly catches these. If a concept relies on them, score it down before delivering.

Generic-photo prompts — "Modern office, professional, 4k, hyperrealistic." Looks like every other AI-stock-photo. INSTEAD: Specific lens, lighting, mood, era, composition, color cast.

Ignoring safe zones — Designing edge-to-edge for Instagram and losing CTA to the crop. INSTEAD: Platform module defines safe zones — concept must fit them.

Treating the brief as a render — Writing a fluffy concept paragraph and calling it done. The brief must be specific enough that a downstream tool/designer can execute without follow-up.

Treating Claude Design output as brand-system input — claude.ai/design generates exploration. It's a starting point downstream of this brief, not a brand commitment.

---

Completion Status Protocol

Per project standard, every run ends with explicit status:

• DONE — brief approved, critic PASS, artifact written
• DONE_WITH_CONCERNS — approved but critic flagged issues; concerns documented in artifact frontmatter. Currently the skill itself ships in this state until platform-modules.md is fully populated.
• BLOCKED — user rejected at a gate, or external dependency missing
• NEEDS_CONTEXT — BRAND.md or DESIGN.md missing; cannot proceed without brand-system

---

Agent Files

Sub-Agent Instructions (agents/)

• agents/brand-anchor-agent.md — Token + sacred element pull
• agents/concept-agent.md — 3 concept directions
• agents/copy-anchor-agent.md — In-asset copy resolution
• agents/brief-synth-agent.md — 3 candidate briefs
• agents/prompt-craft-agent.md — Image-gen prompts
• agents/figma-spec-agent.md — Designer handoff spec
• agents/critic-agent.md — Visual rubric + AI-aesthetic + platform-fit check

Shared References (references/)

• references/asset-types.md — Per-asset format specs
• references/platform-modules.md — Per-platform brief checklists (skeleton — needs follow-up build pass)
• references/prompt-patterns.md — Image-gen prompt structures
• references/visual-rubric.md — Critic scoring dimensions
• references/failure-modes.md — Generic-AI catalog, brand drift
• references/examples.md — End-to-end worked examples