Vibeleaderboard
← Back to Tools
Index / tool

Watch Skill / DeepWatch

github.com/oxbshw/watch-skill
Visit github.com
Category
Developer Tools
Rank
No. 966Tools index
Pricing
Open Source
Platform
cli · web
Type
TOOL
GitHub
355 stars
Latest release
deepwatch-v0.1.4
Date

About

An MIT-licensed toolkit that gives AI agents perception (turning video, audio and screen activity into searchable, timestamped transcripts and OCR text) and deterministic verification (checking file digests, JSON, SQL, HTTP or DOM state via a separate non-LLM process that returns VERIFIED, FAILED, UNVERIFIED or INCONCLUSIVE). It ships as a pip/npm-installable skill with an MCP server (39 tools), CLI and REST API for existing agents, plus DeepWatch, a ready-made agent workspace built on DeepSeek Harness with the skill pre-integrated.

What it can do

  • Transcribe video and audio into timestamped, searchable transcripts

    Video or audio file → Timestamped transcript text

  • Extract text from screen activity via OCR

    Screen activity capture → OCR text

  • Verify file digests through a deterministic non-LLM process

    File → Verification result (VERIFIED, FAILED, UNVERIFIED, or INCONCLUSIVE)

  • Verify JSON state deterministically

    JSON data → Verification result (VERIFIED, FAILED, UNVERIFIED, or INCONCLUSIVE)

  • Verify SQL database state deterministically

    SQL query/data → Verification result (VERIFIED, FAILED, UNVERIFIED, or INCONCLUSIVE)

  • Verify HTTP responses deterministically

    HTTP request/response → Verification result (VERIFIED, FAILED, UNVERIFIED, or INCONCLUSIVE)

  • Verify DOM state deterministically

    DOM/webpage state → Verification result (VERIFIED, FAILED, UNVERIFIED, or INCONCLUSIVE)

  • Expose perception and verification tools to AI agents via MCP server, CLI, and REST API

    Agent requests/tool calls → Tool responses

Why it made the leaderboard

Lets an agent prove it watched a recording and completed a task with timestamped evidence, instead of trusting the model's own claim of what it saw.

Tags

ai-agentsmcpagent-observabilityverificationvideo-analysisocrspeech-to-textdeepseek

Tech Stack

PythonDocker

Comments (0)

No comments yet

Editorially curated, with community endorsements as a secondary signal. Corrections welcome.