Tweet Thread Creator

Create X/Twitter threads for @ExecutionMarket with text-to-image prompts and optional AI-generated images.

Output Structure

Every thread lives under docs/tweets/<thread-name>/:

docs/tweets/<thread-name>/
├── thread.md          # Full thread text (copy-paste ready for X)
├── 1.txt              # Image prompt for post 1
├── 1.png              # Generated image for post 1 (if --generate-images)
├── 2.txt              # Image prompt for post 2
├── 2.png              # ...
├── ...
└── N.txt

Naming rules:

• <thread-name>: kebab-case, short, descriptive (e.g., first-full-flow, openclaw-launch, escrow-deepdive)
• Prompt files: N.txt where N matches the post number in thread.md
• Image files: N.png matching the prompt number
• Thread text: always thread.md

Workflow

Step 1: Understand the Topic

Ask the user (or infer from context):

• What is the thread about?
• What tone? (narrative story, technical breakdown, announcement, hype)
• Any specific transactions, links, or data to include?
• How many posts? (default: 8-12 for a full thread)

Step 2: Write the Thread (thread.md)

Create docs/tweets/<thread-name>/thread.md with:

# Thread Title (internal reference, not posted)

> Description of what this thread covers.

---

## Post 1 (Hook)

Tweet text here...

---

## Post 2 (Beat Name)

Tweet text here...

---
...

Thread writing guidelines:

• Post 1 is always the hook — must grab attention in first line
• Each post should stand alone but advance the narrative
• Include BaseScan links, contract addresses, or data where relevant
• Keep posts under 280 characters when possible (X allows longer with Premium)
• Use line breaks for readability
• End with a closing post that includes https://execution.market
• Tag relevant accounts: @base, @UltravioletaDAO, @ExecutionMarket

Tone for @ExecutionMarket:

• Confident but not arrogant
• Technical but accessible — show, don't lecture
• Story-driven when possible — humans connect with narratives
• Always include verifiable proof (tx hashes, links, contract addresses)

Step 3: Generate Image Prompts (N.txt)

For each post, create a N.txt file with a text-to-image prompt.

Style guide for Execution Market visuals:

• Color palette: deep ultraviolet purple (#7B2FBE), electric blue (#0052FF — Base chain), white, black
• Aesthetic: cinematic, photorealistic base with subtle digital/holographic overlays
• Aspect ratio: 16:9 (default for X timeline)
• No text baked into images (text goes in the tweet itself)
• Mix physical world + digital elements when the story involves both
• Consistent visual language across the entire thread

Prompt structure:

[Main subject]. [Scene/setting details]. [Lighting and atmosphere]. [Color palette notes]. [Style: photorealistic/cinematic/etc]. [Aspect ratio].

Step 4 (Optional): Generate Images (--generate-images)

When the user provides an OpenAI API key and requests image generation:

# Environment variable
OPENAI_API_KEY=<key>

# Generate a single image
python .claude/skills/tweet/scripts/generate_image.py \
  --prompt-file docs/tweets/<thread-name>/N.txt \
  --output docs/tweets/<thread-name>/N.png \
  --size 1536x1024 \
  --model gpt-image-1

# Generate all images for a thread
python .claude/skills/tweet/scripts/generate_all.py \
  --thread-dir docs/tweets/<thread-name>/ \
  --model gpt-image-1 \
  --size 1536x1024

Image generation notes:

• Model: gpt-image-1 (OpenAI's image generation model)
• Size: 1536x1024 for 16:9 aspect ratio
• Reads prompt from N.txt, saves to N.png
• Skips generation if N.png already exists (idempotent)
• Rate limit: ~5 requests/minute, script handles backoff

Usage Examples

Basic (text + prompts only)

User: /tweet
User: Create a thread about the new escrow feature
→ Creates docs/tweets/escrow-feature/thread.md + N.txt files

With image generation

User: /tweet --generate-images
User: Thread about OpenClaw integration
→ Creates thread.md + N.txt + generates N.png via OpenAI API

From existing content

User: /tweet
User: Turn docs/planning/SOME_DOC.md into a thread
→ Reads the doc, creates narrative thread + prompts

Reference: Existing Threads

Important Rules

1. Never auto-post — Only create files. User posts manually.
2. Always create thread.md first, then prompts — thread text drives the visual narrative.
3. One prompt per post — Even if a post has no obvious visual, create a prompt anyway.
4. Keep prompts visual, not textual — No text/numbers baked into images. The tweet provides context.
5. OpenClaw, not OpenCloud — The AI agent platform is called OpenClaw.
6. Ultravioleta branding — Purple (#7B2FBE) is the primary accent. Blue (#0052FF) for Base chain references.