mcpField
C
configuring-vision

configuring-vision

github.com/oxbshwAgent workflows533521 installsUpdated 7w ago
Repository

About configuring-vision

A Watch Skill guide for choosing and configuring a vision provider independently of the agent interface. It serves users of supported agents and REST clients, and explains tiered model routing as well as operation without a vision API.

Key features

  1. 01Configure Anthropic, OpenAI, Gemini, OpenRouter, or Ollama
  2. 02Route bulk work and verification to different models
  3. 03Probe a provider with --verify
  4. 04Keep API keys out of chat
  5. 05Use captions, local transcription, and OCR without vision API

Use cases

  1. 01Set up a vision provider for video analysis
  2. 02Use separate models for scene descriptions and verification
  3. 03Continue evidence search without a vision API

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/oxbshw/watch-skill --skill configuring-vision

After installing, type/configuring-visionin Claude Code to confirm it loaded.

Details

From the project README

Configuring vision

Watch Skill's agent surface and model backend are separate choices. Claude Code, Codex, Cursor, OpenClaw, framework agents, and REST clients all call the same engine; the engine can send selected frames to any supported vision provider.

Supported providers

watch-skill setup-vision --provider anthropic --api-key <KEY>
watch-skill setup-vision --provider openai --api-key <KEY>
watch-skill setup-vision --provider gemini --api-key <KEY>
watch-skill setup-vision --provider openrouter --api-key <KEY>
watch-skill setup-vision --provider ollama

Prefer a key the user already has. Do not claim Ollama is required, and do not ask the user to reveal a secret in chat. They can set the matching environment variable or run the command privately in their terminal.

Route bulk work and verification separately

One model can serve both tiers:

watch-skill setup-vision --provider openai --api-key <KEY> --model <vision-model>

Or use a cheaper model for scene descriptions and a stronger model for uncertain answers and loop critiques:

watch-skill setup-vision --provider openrouter --api-key <KEY> \
  --cheap-model <fast-vision-model> --strong-model <strong-vision-model>

Add --verify to make one live probe call. If it fails, report the structured error and its fix; never echo the key.

No provider is also valid

Without a vision API, Watch Skill still acquires video, reads captions, runs local transcription and OCR, indexes evidence, and searches it. Visual synthesis degrades to timestamped evidence instead of guessing.

Comments

Markdown supported