diff --git a/EvoScientist/skills/image-artist/SKILL.md b/EvoScientist/skills/image-artist/SKILL.md
new file mode 100644
index 0000000..9c8ff28
--- /dev/null
+++ b/EvoScientist/skills/image-artist/SKILL.md
@@ -0,0 +1,62 @@
+---
+name: image-artist
+description: Generate or edit images with the configured dedicated image models. Use when the user asks to draw, illustrate, create a cover/diagram/figure, or modify an existing workspace image.
+---
+
+# Image Artist
+
+Workflow for producing AI-generated images. Execution happens through the
+`generate_image` and `edit_image` tools — this skill tells you when and how
+to call them.
+
+## When to Use
+
+- The user asks to create, draw, or generate an image, cover, illustration,
+  poster, or diagram from a description → `generate_image`.
+- The user provides an existing image (in `artifacts/` or another workspace
+  path) and asks to modify, restyle, or adapt it → `edit_image`.
+- The user asks for a chart/plot of **data** → do NOT use this skill; compute
+  the plot with code (matplotlib) instead.
+
+## Prompt Construction
+
+Write the `prompt` as a single dense paragraph covering:
+
+1. **Subject** — what is in the image, concretely.
+2. **Style** — e.g. flat vector, watercolor, photorealistic, line art.
+3. **Composition** — layout, perspective, framing.
+4. **Palette/lighting** — only when the user cares.
+5. **Negative constraints** — what to avoid (text artifacts, extra limbs),
+   phrased as "no ..." at the end.
+
+Do not pad with fluff ("masterpiece", "best quality"); describe the image.
+
+## Parameters
+
+- `size`: "1024x1024" (default), "1536x1024" (landscape), "1024x1536"
+  (portrait). Pick from the content: covers/posters often landscape or portrait.
+- `model`: omit to use the configured default; the tool description lists the
+  available image models. Only pass a name from that list.
+- `output_path`: suggest a meaningful name like `artifacts/topic_cover.png`;
+  must start with `artifacts/` and end with `.png`.
+- `n`: number of variations (default 1). For "give me a few options" use 2-4.
+
+## After the Call
+
+- The tool returns JSON: `{"ok": true, "paths": [...]}` or
+  `{"ok": false, "error": "..."}`.
+- On success, reference the saved paths in your reply.
+- On failure, read the error: if it lists available models, retry with one of
+  them or explain the configuration gap to the user. If the provider rejected
+  the prompt, rewrite it more concretely and retry once.
+
+## Variations
+
+For "make variations", call `generate_image` once with `n>1`, or call
+`edit_image` on a previously generated image with a modified prompt.
+
+## Security Rules
+
+- Never read, print, or quote the `image_generation` section of config.yaml
+  or any API key. The tools handle credentials internally.
+- Never pass file paths outside the workspace as `image_path`/`mask_path`.