niyongsheng/free-vision-skill
Local‑only vision skill for macOS Localized image recognition skill
Fully-local image understanding via macOS Vision: view_image (describe scene/people/QR/composition), ocr_image (extract text; layout=true for table structure + coordinates), with input as http(s) URL / base64 / local path. Includes paste-to-path for the Web UI: ⌘V an image in the DSH web input box inserts its local absolute path (loopback-only upload, magic-byte checked for PNG/JPEG/GIF/WebP/HEIC/HEIF). Zero cloud calls — OCR, table extraction, and description all run on-device.
Install
dsh plugin add @niyongsheng/free-vision-skilldsh plugin add @niyongsheng/free-vision-skill (npm 0.3.1 verified on registry), then add the insert row to cordis.patch.yml: - insert: - id: free-vision-skill, name: '@niyongsheng/free-vision-skill', config: { timeout: 120000 }. macOS only: requires macOS 11+ and Xcode Command Line Tools; first run compiles ~5–10s, cached afterwards. Also usable as a Claude Code skill (swift scripts/ocr.swift ...).
Compatibility
macOS 11+ only (uses Apple Vision framework); fully local — images never leave the Mac. DSH plugin or Claude Code skill.
Details
- Repo: niyongsheng/free-vision-skill
- Category: Agent Capabilities
- Stars: 3
- Version: npm 0.3.1 (registry-verified 2026-08-29)
- Last push: 2026-08-19
- First seen: 2026-08-08
Recent updates
macOS Vision OCR/describe tools; fully local; paste-to-path in Web UI; layout mode for tables.
FAQ
- Does it upload images anywhere?
- No — it is fully local via the macOS Vision framework; images never leave your Mac.
- What tools does it register?
- view_image (describe) and ocr_image (text; layout=true for table structure + coordinates).
- Which platforms are supported?
- macOS 11+ with Xcode Command Line Tools. Also usable outside DSH as a Claude Code skill.
Alternatives
FuzzySoul/dsh-free-vision · 54xkeee/dsh-vision · tiefeiyu/dsh-see-image