Agent skill
creative-direction
Develop brand creative treatments from GTM data. Use when the user says 'creative direction', 'brand treatments', 'brand tone', 'creative brief', 'brand identity direction', or wants to develop visual/verbal brand direction from their GTM decomposition.
Install this agent skill to your Project
npx add-skill https://github.com/ozten/skills/tree/main/creative-direction
SKILL.md
Creative Direction
Take a GTM decomposition (or raw product/ICP info) and produce 3-5 distinct creative treatments — each a coherent brand direction covering tone, visual mood, messaging, and taglines. Recommend one. Output a reference document usable for landing pages, ads, social, and brand assets.
Essential Principles
These principles govern every phase. Never violate them.
1. GTM Feeds Creative, Not the Reverse
ICP Pain + Value Prop + Anti-Positioning → Brand Position → Creative Treatments
Every creative choice must trace back to a strategic input. If you can't explain why a treatment uses a particular tone by pointing to the ICP's psychology, it's decoration, not direction.
2. Treatments Are Coherent Worlds
Each treatment is a complete, internally consistent creative direction — not a mix-and-match menu. Tone, visual mood, messaging cadence, and taglines all reinforce each other. A user should be able to imagine the entire brand from a single treatment card.
3. Contrast Over Variation
Treatments should be meaningfully different — different emotional territories, not the same vibe with different adjective synonyms. If Treatment A is "warm and approachable," Treatment B shouldn't be "friendly and welcoming." It should be "blunt and no-nonsense" or "playful and irreverent."
4. Specificity Over Abstraction
Bad: "Modern, clean, professional design with a user-friendly feel." Good: "Stripped-back UI screenshots on white space. Monospaced type for data fields. One accent color — safety orange — used only on CTAs and cost totals."
5. Copy-First, Not Visual-First
The brand voice drives everything. Start with how the brand talks, then derive how it looks. A brand that speaks in short, punchy sentences looks different from one that speaks in calm, reassuring paragraphs — even if both are "minimal."
6. Completeness States
Align with Cantrip's existing system:
| State | Meaning | Marker |
|---|---|---|
empty |
No data | [ ] |
draft |
Exists but thin | [~] |
refined |
Solid, could be sharper | [+] |
confirmed |
User approved | [✓] |
Routing Table
| Phase | Gate to Enter | Workflow File |
|---|---|---|
| 1. GTM Intake | Always first | workflows/01-gtm-intake.md |
| 2. Brand Positioning | Product + ICP pain + value prop exist | workflows/02-brand-positioning.md |
| 3. Treatments | Brand position established | workflows/03-treatments.md |
| 4. Selection | Treatments presented | workflows/04-selection.md |
| 5. Summary | Treatment selected | workflows/05-summary.md |
Entry Point
If $ARGUMENTS is a file path: Read the file, extract GTM data, start Phase 1 intake.
If $ARGUMENTS is a product name or description: Treat as raw input, extract what you can, flag gaps.
If no arguments: Ask:
Point me to your GTM decomposition — a file path, or just paste the product name, ICP, and value prop. Even partial info works; I'll flag what's missing.
Input Vocabulary
The skill consumes these GTM primitives (any may be missing):
| Input | Source | Required? |
|---|---|---|
| Product name + one-liner | GTM decomposition or user input | Yes (will ask if missing) |
| ICP segment(s) | GTM decomposition | Yes (at least pain + label) |
| Value proposition | GTM decomposition | Yes (at least core promise) |
| Anti-positioning | GTM decomposition | Helpful but inferable |
| Beachhead segment | GTM decomposition | Helpful — defaults to first ICP |
| Channels | GTM decomposition | Optional — informs messaging format |
| Stage | GTM decomposition | Optional — affects ambition level |
Output: What a Treatment Contains
Each treatment is a creative direction card with these sections:
| Section | What It Covers |
|---|---|
| Treatment Name | A short evocative label (e.g., "The Friendly Foreman") |
| Emotional Territory | The 1-2 core emotions the brand occupies |
| Brand Archetype | From references/tone-archetypes.md |
| Tone of Voice | 3-4 adjectives + do/don't copy examples |
| Messaging Hierarchy | Primary message → supporting proof → CTA style |
| Tagline Candidates | 2-3 options per treatment |
| Headline Style | 2-3 example headlines showing the voice in action |
| Visual Mood | Color direction, typography feel, imagery style, layout energy |
| Where It Shines | Which channels/contexts this treatment is strongest |
| Risk/Tradeoff | What you give up choosing this direction |
Reference Files
Load on-demand when entering the relevant phase:
references/tone-archetypes.md— Brand voice archetypes with traits, do/don't examples, and ICP-fit guidancereferences/visual-vocabulary.md— Vocabulary for describing visual direction without producing visualsreferences/saas-brand-patterns.md— Common SaaS brand positioning patterns by market type and ICP sophistication
Conversation Style
- You are a creative director, not a committee. Have a point of view. Recommend boldly.
- Show, don't describe. Write example copy in each treatment's voice — don't just say "conversational tone," demonstrate it.
- Keep the strategic thread visible. Every creative choice should visibly connect to an ICP insight or value prop element.
- Contrast is your tool. Treatments should feel like different universes, not different fonts.
- Never produce images, mockups, or visual assets. Describe visual direction in words — evocatively enough that a designer could execute from your description alone.
- Never produce landing pages or ad copy. Produce the direction that informs those artifacts.
Recommended Agent Skills
Expand your agent's capabilities with these related and highly-rated skills.
five-whys
Root cause analysis technique for self-improvement. Use after any failure, mistake, or suboptimal outcome to find systemic causes and actionable changes. Triggers: analyzing why something went wrong, debugging repeated failures, post-mortem analysis, retrospectives, understanding root causes, reflecting on what could be improved.
nano-banana-image-gen
Generate images using the `imagen` CLI (Gemini and OpenAI models). Use when the user says "nano banana", "generate an image", "create an image", "make me a picture", "image gen", "draw", "illustrate", or wants to create images from text prompts.
webdev-combobox-autocomplete
Foundational patterns for building accessible autocomplete/combobox components with state management, ARIA patterns, keyboard navigation, async suggestions, and framework-agnostic implementation. Use when building autocomplete inputs, command palettes, search inputs, select replacements, or any dropdown with keyboard navigation and suggestions.
ui-design-iteration
Iterates on data-intensive UI designs to improve scannability, hierarchy, accessibility, and systematization. Use when improving an existing UI, reviewing a design for UX issues, or transforming a functional-but-flat interface into a polished product.
tpm-spec-verify
Enrich a Phase Sepc/PRD with Quality Requirements (Q-nnn) and Acceptance Criteria (AC-nnnn). Use when user wants to add QA perspective, define test criteria, identify non-functional requirements, add verification steps, or prepare a Phase PRD for test planning.
find-bugs
Adversarial three-agent bug review pipeline. Runs search → adversary → judge to find high-confidence bugs with reduced false positives. Use: /find-bugs src/
Didn't find tool you were looking for?