oxbshw/watch-skill

Give AI agents eyes, ears, and verifiable results. Watch Skill turns video, audio and screen activity into searchable, timestamped evidence and proves work with deterministic contracts, not model opinion. DeepWatch is the agent workspace built on DeepSeek Harness. Python + npm, MCP, CLI, REST, Web.

What it solves

Watch Skill provides AI agents with the ability to "see" and remember video content. It solves the problem of agents needing to re-process videos every time a question is asked and the lack of reliable verification for agents performing tasks in a browser or on a desktop, where a visual "success" message might hide a technical failure.

How it works

The system creates a searchable, timestamped index of videos, live streams, and screen recordings using scene-aware frames, on-screen text (OCR), and transcription. It uses a hybrid retrieval system (full-text and vector) to answer questions with specific timestamp citations. For agentic workflows, it implements "THE LOOP," a capture-critique-fix cycle that records an agent's actions and verifies them against deterministic contracts and network requests to ensure the task actually succeeded.

Who it’s for

It is designed for developers building AI agents (using frameworks like LangChain, CrewAI, or AutoGen) and users of AI coding assistants (such as Claude Code, Cursor, or Windsurf) who need their agents to analyze video evidence or operate browsers with high reliability.

Highlights

  • Persistent Video Memory: Analyze a video once and perform multiple lookups without re-downloading or re-transcribing.
  • THE LOOP Verification: A rigorous verification process for browser/desktop flows that combines visual critique with deterministic checks (e.g., checking if a network request actually returned a 200 OK).
  • Local-First Processing: Supports local Whisper for transcription and Ollama for vision, reducing dependency on cloud APIs.
  • Wide Integration: Ships as MCP tools, Claude Code skills, and native adapters for major agent frameworks.
  • Evidence-Based Answers: Provides timestamped citations and confidence levels rather than generic summaries.

Related

  • Project
    octimot/StoryToolkitAI

    An AI-powered film editing assistant that transcribes, indexes, and searches video footage, allowing editors to use LLMs to automatically create story selections for export to editing software.

  • Project
    microsoft/skill-recorder

    A tool that records user screen activity and uses GitHub Copilot to convert those demonstrations into reusable skills and automations for AI agents.

  • Project
    Memento-Teams/Memento-Skills

    A skill-centric agent runtime that allows AI agents to design, execute, and persist reusable skills to improve their performance on complex tasks through a reflection-based learning loop.

  • Project
    Goldentrii/AgentRecall-X

    A persistent memory system for AI agents that tracks and measures whether human corrections actually change agent behavior across sessions to prevent repeating mistakes.

  • Project
    nossa-y/activity-frames

    activity‑frames records your screen locally, compiles raw snapshots into deterministic “activity frames”, and exposes them as compact context blocks or executable scripts for AI agents, dramatically reducing token cost and improving reliability of repeated computer‑use tasks.