Video Perception agent plugin
This plugin gives a shell-capable agent practiced video perception over the
public video-perception CLI and contracts. It does not include a second
processing engine or install dependencies.
The agent learns to:
- apply the bundled coarse-to-fine video perception playbook;
- discover this machine's existing video capabilities;
- request the smallest useful Strategy or SensePlan;
- create and query provenance-bearing Observations;
- author decentralized Provider and Knowledge Pack manifests.
video-perception capabilities --json
video-perception doctor --json
video-perception catalog list --json
video-perception extension list --json
Skills are perceive-video, add-video-provider, and
create-perception-pack. Bundled strategies and the foundation knowledge pack
live beside the Skills under this plugin directory.
Install
The video-perception executable must already be on PATH; the plugin
installs nothing. Hosts that load MCP servers from a plugin start
video-perception mcp from mcp.json (Agent Plugins) or .mcp.json (Claude
Code).
Claude Code. The repository is a marketplace:
claude plugin marketplace add opus-pro/video-perception
/plugin install video-perception --marketplace opus-pro/video-perception
To work on the plugin, validate and load the directory directly:
claude plugin validate plugin
claude --plugin-dir /absolute/path/to/video-perception/plugin
Codex and ChatGPT. The same repository is a Codex marketplace. The
portable plugin.json carries the OpenAI presentation under
extensions.com.openai; .codex-plugin/plugin.json stays as the
compatibility fallback for older builds.
codex plugin marketplace add opus-pro/video-perception
Then install video-perception from the plugin browser.
Any Agent Plugins 1.0 client. Point the client at this directory.
plugin.json, mcp.json, and skills/ are the portable components;
.claude-plugin/ and .codex-plugin/ are client overlays that other clients
ignore.
DeepSeek Harness uses the separate thin adapter under
integrations/deepseek-harness.