generate-agent-graphics

name	generate-agent-graphics
description	Create consistent, memorable visual graphics for AgentWorkforce agents in the agents repo. Use when asked to generate, design, refresh, or wire README graphics, avatars, badges, cards, banners, or visual identity assets for folders such as granola, hn-monitor, linear, repo-hygiene, review, spotify-releases, and vendor-monitor.

Generate Agent Graphics

Goal

Create a recognizable family of bitmap graphics for each deployable agent while keeping every agent visually distinct. Prefer generated PNG assets over Canva or SVG unless the user explicitly asks for Canva or vector output.

Output Contract

For each target agent folder, create:

• avatar.png — square 1024x1024 hero icon, recognizable at small size.
• card.png — 16:9 README/social card using the same metaphor.
• banner.png — 1400x392 README banner crop, matching the relay README image format: full width, not too tall.

Do not put text, agent names, logos, watermarks, or fake UI copy inside the generated image. Add labels in Markdown, not in pixels.

Visual System

Use one shared style for all agents:

Polished 3D editorial icon, dark graphite studio background, luminous glass and brushed-metal object, one distinctive accent color, clean studio lighting, centered composition, high contrast, generous padding, no text, no logo, no watermark.

Use distinct metaphors and accent colors:

Agent	Metaphor	Accent
granola	meeting transcript pages becoming a feature spark and PR branch	warm gold
hn-monitor	radar dish scanning floating news/story cards	orange
linear	task ticket moving through a conveyor into a PR branch	blue-violet
repo-hygiene	microscope over codebase blueprint, duplicate paths, pruned dead branches	teal
review	shield, code diff checklist, green CI lights	emerald
spotify-releases	glowing vinyl record emitting release pulse waves	neon green
vendor-monitor	package cube with orbiting dependency nodes and radar beam	amber

Read references/agent-prompts.md for ready-to-use prompts and prompt rewrite rules.

Workflow

1. Identify target agent folders from the user request. If none are specified, use every top-level folder with both persona.ts and agent.ts.

2. Generate avatar.png first for each agent using the built-in image generation tool. Use one image-generation call per agent so the metaphor can be specific.

3. Generate card.png for each agent using the same metaphor and accent color, but request a 16:9 README/social banner composition.

4. Create banner.png from card.png as a centered 1400x392 crop so README images fill the content width without becoming too tall.

5. Save final selected files into the agent folders. Never leave project assets only in $CODEX_HOME/generated_images.

6. Update READMEs to display the banner near the top using the same pattern as the relay README. Use:

   <img src="./banner.png" alt="<Agent Name>">
   

7. Run the README helper if useful:

   python .agents/skills/generate-agent-graphics/scripts/apply_readme_images.py .
   

8. Check git status --short, then report changed paths and any assets that still need regeneration.

Quality Bar

Reject and regenerate any image that:

• contains readable or pseudo-readable text
• looks like a generic robot instead of the agent metaphor
• lacks a clear focal object at small size
• uses a different global style from the rest of the set
• has watermarking, brand logos, UI screenshots, distorted hands/faces, or busy collage composition

Canva

Use Canva only as an optional polish layer after repo-native assets exist. If the user asks for Canva, create/export designs from a reusable template, but keep the final PNG files committed in each agent folder as the source consumed by READMEs.