vision · kotot — Claude Code skill: what it does, how to install
Sends images to an external vision model when your coding agent lacks native image understanding. Handles OCR, screenshots, and diagrams via a local helper script.
npx skills add kotot/vision --skill vision -a claude-codeWhat vision does
This skill bridges the gap when an agent cannot view images directly. It supplies a bundled Python script, scripts/see.py, which packages the target image and your prompt into an API call. The external model then returns a plain text response to standard output.
The script relies solely on the Python standard library without extra dependencies. Setup requires two environment variables: VISION_BASE_URL pointing to the full chat completions endpoint and VISION_MODEL indicating the chosen model. You can store them in shell variables or in .claude/settings.json.
Local server setups can skip VISION_API_KEY. If you need to verify what gets sent, run the script with --dry-run to inspect the payload. Without a reachable completions endpoint, every execution fails immediately.
What it helps with
- extract text and tables from scanned documents or receipts
- diagnose errors shown in terminal or browser screenshots
- interpret architecture diagrams and user interface mockups
- convert chart values into structured markdown tables
Example prompts for the agent
- Inspect error.png and explain why the build failed
- Transcribe the text from receipt.png preserving the original structure
- Examine diagram.png and describe the components shown
How to install vision
1Skills CLI (Vercel)
npx skills add kotot/vision --skill vision -a claude-code2Manually: put the files in place
~/.claude/skills/vision/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/vision/
1Skills CLI (Vercel)
npx skills add kotot/vision --skill vision -a codex2Manually: put the files in place
.agents/skills/vision/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add kotot/vision --skill vision -a cursor2Manually: put the files in place
.cursor/skills/vision/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install kotot/vision vision2Manually: put the files in place
.github/skills/vision/SKILL.md1Skills CLI (Vercel)
npx skills add kotot/vision --skill vision -a gemini-cli2Platform CLI
gemini skills install https://github.com/kotot/vision --path visionManually: put the files in place
.devin/skills/vision/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add kotot/vision --skill vision -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about vision
How to install vision in Claude Code?
Run npx skills add kotot/vision --skill vision -a claude-code in your terminal.
What does the vision skill do?
It delegates visual inspection tasks to a configured vision model using a Python script. The script converts images to base64 when necessary and outputs the model's text answer to stdout.
What configuration is required before running the script?
You must define VISION_BASE_URL with the complete chat completions endpoint and set VISION_MODEL. These values belong in your environment variables or within .claude/settings.json.
Similar skills
All skills for Agent setup: memory, context, orchestration →vision by other authors comes first.
