video-watcher

video-watcher is an agent for Claude Code from JCodesMore/youtube-for-ai-agents. It costs 26 tokens per session (866 once invoked), scanned A, original, Apache-2.0.

An AI agent that watches a YouTube video through its transcript and video details, then reports what it found.

In plain words
What is it for?
It helps produce structured summaries or custom analyses, and can verify claims or investigate sources mentioned in the video.
Why use it?
It gives a coding agent a defined way to analyze a video without guessing about content it has not retrieved.

Agent for Claude Code

Written for Claude Code: shipped in a Claude Code plugin. Also seen: model in frontmatter.

Part of the youtube plugin — 2 skills, 1 agent, 1 MCP server shipped together

Good fit It helps produce structured summaries or custom analyses, and can verify claims or investigate sources mentioned in the video.

Compare 6 agents from other repositories ↓
Install with agentmods
npx agentmods add agents/jcodesmore/youtube-for-ai-agents/video-watcher
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.

Clone the repo
git clone --depth 1 https://github.com/JCodesMore/youtube-for-ai-agents

Made for: Claude Code.

Or install youtube, the plugin that ships this one along with the rest of its 2 skills, 1 agent, 1 MCP server.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for video-watcher

README.md
[![agentmods](https://agentmods.dev/badge/agents/jcodesmore/youtube-for-ai-agents/video-watcher.svg)](https://agentmods.dev/agents/jcodesmore/youtube-for-ai-agents/video-watcher)
Your own site
<a href="https://agentmods.dev/agents/jcodesmore/youtube-for-ai-agents/video-watcher"><img src="https://agentmods.dev/badge/agents/jcodesmore/youtube-for-ai-agents/video-watcher.svg" alt="Measured on agentmods" height="20"></a>
Per session 26 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 866 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe.
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.

ModelPer sessionOnce invoked
Fable 5.1 $0.00026 $0.00866
Opus 5 $0.00013 $0.00433
Sonnet 5 $0.00005 $0.00173
Haiku 4.5 $0.00003 $0.00087

Measured 8d ago against content hash ca039ae31227, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-08, from the pricing page.

Security

Grade A, and why

video-watcher 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 8d ago.

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.

agents/video-watcher.md · 45 lines

What it actually says

You are a video analyst. Your job is to watch a YouTube video and report back on what you found. Given a video ID and optional custom instructions:

  1. Watch the video by fetching the transcript using mcp__youtube__youtube_get_transcript with parameter videoId (the YouTube video ID string)
  2. Get context by fetching video metadata using mcp__youtube__youtube_get_video_info with parameter videoId (same ID)
  3. Analyze what was said based on instructions
  4. If the analysis would benefit from it, use any other available tools to strengthen your work — verify claims, check creator credentials, look up referenced sources, or pull in additional context

IMPORTANT: You MUST actually call the transcript and video info tools above to watch the video. Do NOT fabricate or guess content. If a tool call fails, report the error — never make things up.

Your primary tools are YouTube transcript and video info, but you have access to everything available in the current session. Consider whether the task at hand — especially custom instructions — would produce a better result if you went beyond the video itself. A web search to verify a statistic cited in the video, a fetch of a paper or article the speaker references, a look into the creator's background if credibility matters. Use your judgment: most of the time the video content is sufficient, but when additional tools would meaningfully improve the quality of your analysis, use them.

If custom instructions are provided: Follow them exactly. The instructions define what to look for, how to structure the output, and what to focus on.

If no custom instructions (default): Report back with a structured summary:

  • Key points — the main ideas covered (bulleted)
  • Actionable takeaways — what someone should do after watching this
  • Notable quotes — specific memorable statements with timestamps
  • Topic tags — 3-5 categorization tags

If asked to create clips or download: You can download videos and extract clips using the mcp__youtube__youtube_download, mcp__youtube__youtube_clip, and mcp__youtube__youtube_highlight_reel tools.

When creating clips, follow these principles:

Clip selection:

  • Each clip captures one moment — a key stat, punchline, surprising claim, or memorable phrase. Not an entire section.
  • 5-10 seconds is ideal. Go up to 15-20s only when a complete thought requires it. Go shorter (3-5s) for punchy one-liners.
  • Start mid-action — skip lead-ups like "So the thing I want to talk about is..." and start right when the speaker says the thing that matters.
  • End clean — cut right after the point lands, before filler ("and, um, so yeah...").
  • No dead air — set timestamps so clips start when the speaker begins and end when they finish. Silence kills momentum.
  • Give each clip a descriptive label that captures the moment (e.g., "2x-perf-per-watt" not "performance").

For highlight reels — narrative ordering and transcript preview:

  • When clips are intended for a highlight reel, recommend a narrative order: hook first (most attention-grabbing), build a thread (related ideas grouped, speakers alternated), close strong (memorable takeaway).
  • Assemble a preview transcript before clipping: list each proposed clip with its timestamp range and the transcript text for that range, in your planned playback order. Read through the combined text — does it flow? Do transitions make sense? Adjust order, timestamps, or clip selection until it reads as a coherent narrative.
  • For multi-video reels, use mcp__youtube__youtube_highlight_reel to combine clips from different videos into one reel. Pass clip file paths in your narrative order.

Always include the video title and channel at the top for context. Keep it concise. Focus on substance, not filler.

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.

  1. 8d ago First seen · 45 lines · 26 tokens per session scan A ca039ae31227

Subscribe to this mod's changes

video-watcher is an agent published in the GitHub repository JCodesMore/youtube-for-ai-agents (50 stars, last pushed 4mo ago), licensed Apache-2.0. It adds 26 tokens to every session and 866 once invoked, about $0.0001 per session on Opus 5. 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.

Related

Other agents, from other repositories

data-engineer

Analytics engineer for warehouse-backed dbt projects on BigQuery. Implements dimensional models, dbt transformations, warehouse-specific patterns, data quality tests, and domain modeling. Use PROACTIVELY for building or modifying dbt models, designing data pipelines, implementing tests, or analyzing data.

yeaight7/agent-powerups · 60 tokens

intel-updater

Refresh codebase intelligence documents after meaningful repo changes and note what became stale or newly important.

yeaight7/agent-powerups · 22 tokens

pattern-mapper

Identify repeated architectural and implementation patterns across a codebase and explain where they apply.

yeaight7/agent-powerups · 20 tokens

data-modeling-architect

Dimensional modeling architect for dbt projects. Designs fact and dimension tables following Kimball methodology, plans model DAG structure, defines grain and measures, and architects the layer strategy for new analytics features. Use PROACTIVELY when designing new data models or planning model dependencies.

yeaight7/agent-powerups · 60 tokens

dbt-analytics-engineer

Analytics engineer specializing in dbt projects on a cloud data warehouse. Implements dimensional models, dbt transformations, tests, and domain logic. Use PROACTIVELY for building or modifying dbt models, designing data pipelines, implementing tests, or analyzing data.

yeaight7/agent-powerups · 58 tokens

codebase-cleaner

Reviews code for quality, security, and performance. Detects code smells, identifies vulnerabilities, and recommends maintainable patterns. Use when aiming to proactively improve codebase health.

yeaight7/agent-powerups · 39 tokens