A Claude Code integration for watching and analysing online videos. It can process a video into scene-aware images, text read from those images, captions, and searchable follow-up information.

Use it to inspect YouTube videos, extract frames and captions, recognise text in scenes, and ask follow-up questions about an indexed video.

It removes the need to watch a long video manually when you need to find scenes, read on-screen text, or search its content. The supplied example shows it processing a seven-minute YouTube video inside Claude Code.

claude-code — Agent by oxbshw

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods

npx agentmods add agents/oxbshw/watch-skill/claude-code

Clone the repo

git clone --depth 1 https://github.com/oxbshw/watch-skill

Per session 0 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.

When invoked 517 The whole file, excluding the scripts and references it only reads on demand.

Security scan A 0 findings. Scan, not verified.

Origin original No closer match found in the catalogue.

Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

Model	Per session	Once invoked
Fable 5	$0.00000	$0.00517
Opus 5	$0.00000	$0.00259
Sonnet 5	$0.00000	$0.00103
Haiku 4.5	$0.00000	$0.00052

Measured yesterday against content hash 1e066956a683, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

claude-code scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

docs/agents/claude-code.md · 62 lines

What it actually says

Watch Skill in Claude Code

Status: machine-tested ✅ (this repo's MCP server runs registered in Claude Code on the development machine; all tools exercised end-to-end. Last live run 2026-07-06: the /watch skill adapter processed a real 7-minute YouTube video inside a Claude Code session — doctor green, 25 scene-aware frames, OCR on 24, captions transcript, indexed for follow-ups — with no manual intervention).

Install

uvx --from "watch-skill[standard]" watch-skill doctor   # bootstraps ffmpeg + yt-dlp

git clone https://github.com/oxbshw/watch-skill && cd watch-skill
uv sync --extra all
uv run watch-skill doctor

Register the MCP server

claude mcp add watch-skill -- uvx --from "watch-skill[standard]" watch-skill serve

Or per-project via .mcp.json in your project root:

{
  "mcpServers": {
    "watch-skill": {
      "command": "uvx",
      "args": ["--from", "watch-skill[standard]", "watch-skill", "serve"]
    }
  }
}

pip install instead of a checkout? Use "command": "watch-skill", "args": ["serve"].

Smoke test (3 steps)

Restart Claude Code, then run /mcp — watch-skill should be listed as connected.
Ask: "Use watch-skill to watch https://www.youtube.com/watch?v=aqz-KE-bpKQ and tell me what happens at 0:10."
Ask a follow-up: "Ask the same video what the bird is doing" — it should answer via ask_video in seconds, without re-downloading.

Notes

Tool responses include frames as real images (Claude sees them inline), capped at 12 per response.
The Claude Skill adapter (``) offers /watch as a slash command on top of the same engine — see that README.

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

yesterday First seen · 62 lines · 0 tokens per session scan A 1e066956a683

Subscribe to this mod's changes

claude-code is an agent published in the GitHub repository oxbshw/watch-skill (320 stars, last pushed yesterday), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 517 tokens. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.

Research Testing Writing and docs agent-tools ai-agents claude claude-code computer-vision ffmpeg

project-auditor

Read-only health audit of the VibeFrame monorepo. Use for a periodic project review — structure, CLI/doc drift, dead code, dependency hygiene, test/lint health, and tech debt — producing a prioritized cleanup plan. Proposes actions; does not make sweeping edits.

vericontext/vibeframe · 62 tokens

claude-code

Getting it into your agent

What it costs to keep this loaded

Grade A, and why

Nothing flagged

What it actually says

Watch Skill in Claude Code

Install

Register the MCP server

Smoke test (3 steps)

Notes

What this file has done since we first saw it

Other agents, from other repositories

project-auditor

AGENTS.motiscope

qa-reviewer

custom_agent

domain

query_flow_agent