YouTube Skills for AI Agents
Give AI agents YouTube transcript retrieval, video search, channel browsing, and playlist extraction — without yt-dlp or headless browsers.
- Source repo
- ZeroPointRepo/youtube-skills
- Stars
- ★ 1.1k
- Last updated
- 2d ago
- License
- MIT
- FA score
- 47/100 · Major gaps
At a glance
- How it runs
- Works with
- Portable with changesCodex · Claude Code
- Cost
- Free tier plus a paid hosted plan
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- Researchers ask an agent to fetch and summarize transcripts of the top 5 videos on a topic
- Not a fit if
- Teams that need fully offline or on-prem data handling; every data call goes through TranscriptAPI's cloud
- Users unwilling to depend on a third-party commercial API that can change terms or go down
- Anyone needing caption-less, private, or still-live videos, which return 404
- Source review
- 47/100 · Major gaps 1 safety controls not found
What does this agent do, and when should you use it?
This repository ships a set of Agent Skills-format skill packages whose core product, the youtube-full skill, wraps the TranscriptAPI.com REST API to fetch timestamped YouTube transcripts, search videos (up to 50 results), resolve @handles and browse channel uploads, and extract all videos from playlists. Installation runs through npx skills add, npx clawhub, or hermes skills install; 12 skills ship in total, with 5 core skills and the rest as narrower variants or aliases. On first run, the agent walks the user through TranscriptAPI signup and saves the API key automatically (100 free credits). It avoids yt-dlp, headless browsers, and Google YouTube Data API keys, but every data operation depends on TranscriptAPI's quota, making it a clearly third-party-service-bound tool.
Once installed, the agent invokes skills from plain-language instructions: the skills call TranscriptAPI REST endpoints to retrieve timestamped video transcripts, up to 50 search results (titles, thumbnails, view counts, publish dates), paginated channel uploads (100 per page), and playlist contents, and can bulk-extract transcripts for every video in a channel or playlist. The API key is stored in the TRANSCRIPT_API_KEY environment variable or runtime-specific locations (~/.openclaw/openclaw.json, the Hermes secret store, shell config files, with ~/.transcriptapi as fallback). Channels accept @handle, a channel URL, or a UC ID; playlists accept PL, UU, LL, FL, and OL prefixes.
- Researchers ask an agent to fetch and summarize transcripts of the top 5 videos on a topic
- Content teams bulk-generate transcripts for a channel's uploads from the past week
- Analysts extract captions (e.g., in Spanish) from every video in a playlist
- Developers who want YouTube search inside an agent without a Google YouTube Data API key
- OpenClaw or Hermes Agent users querying the latest videos from channels like TED or NASA conversationally
How do you install or deploy this agent?
OpenClaw (ClawdBot/Moltbot): npx clawhub@latest install youtube-full
Hermes Agent: hermes skills install skills-sh/ZeroPointRepo/youtube-skills/skills/youtube-full
Claude Code / Cursor / Antigravity / Cline / Codex: npx skills add ZeroPointRepo/youtube-skills --skill youtube-full
Manual: git clone https://github.com/ZeroPointRepo/youtube-skills.git and copy skills/youtube-full into ~/.claude/skills/On first run, the agent auto-registers with TranscriptAPI (email → OTP from inbox → key saved automatically); alternatively sign up manually at transcriptapi.com, copy your sk_ key, and export TRANSCRIPT_API_KEY="sk_your_key_here".
How do you use this agent?
After installing youtube-full, just ask the agent in plain language: "Summarize this video: [URL]" (fetch transcript and summarize), "Find videos about machine learning" (search), "List all videos from the TED channel" (browse channel), or "Get transcripts for all videos in this playlist: [URL]" (bulk extraction). Channel resolve and latest-updates operations are free; most other operations cost 1 credit each. The free tier is rate-limited to 300 requests/minute.
What are this agent's strengths and limitations?
- Explicit first-class support for OpenClaw, Hermes Agent, Claude Code, Cursor, Antigravity, Cline, Codex, and more, each with a native install command
- No yt-dlp, headless browsers, or Google Cloud project/OAuth setup — sidesteps cloud-IP blocking and quota hassles
- 12 skills split by function; install only transcript or youtube-search for a smaller context footprint
- 12 skills, a 100-credit free tier, and a 300 req/min rate limit make onboarding cheap
- All data features depend on the third-party commercial TranscriptAPI; beyond the 100 free credits you pay ($5/month Starter or $54/year annual)
- Vendor concentration risk: TranscriptAPI is an independent product unaffiliated with YouTube/Google, so its terms or availability directly affect the skills
- Transcript availability is limited: 404 for caption-less, private, or still-live videos, and many videos only carry captions in their original language
- A 403 requires sending a specific User-Agent header (e.g., HermesAgent/0.11.0) and 422 enforces strict channel/playlist reference formats, raising the debugging bar
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| YouTube Skills for AI Agents This agent | 47 · Major gaps | Agent plugin / skillFreemium | ★ 1.1k | 2d ago | — | Codex · Claude Code |
| Pydantic AI Skills | 86 · Good | Library / SDKFree + model costs | ★ 380 | 7d ago | Python | OpenAI API |
| Finding-Unknowns Skills | 76 · Good | Agent plugin / skillFree | ★ 353 | 25d ago | Python | Claude Code · Claude.ai |
| Self-Learning Skills | 56 · Major gaps | Agent plugin / skillFree + model costs | ★ 970 | 1mo ago | — | Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence shows: SECURITY.md states skills never embed credentials, keys are runtime-only via environment, and third-party service disclosure is present. Deductions: the automated setup flow has the agent register the user's email with TranscriptAPI, relay an OTP, and write the key into multiple shell config files (.zshenv/.bashrc etc.), a broad footprint with no documented per-step confirmation; no SKILL.md files are present to verify the least-privilege claim; full dependence on an unverified paid third-party API with external data egress to transcriptapi.com not itemized at skill level; no rollback/uninstall guidance.
Evidence shows: a thorough troubleshooting table covering 401/402/403/404/408/422/429 with causes and fixes. Deductions: no SKILL.md or code in the reviewed file set to verify instruction self-consistency or endpoint behavior; hard dependency on a single third-party service with no degraded mode or local fallback, so availability is bound entirely to an external SLA.
Evidence shows: install commands for multiple runtimes (OpenClaw, Hermes, Claude Code, Cursor, etc.), a task/prompt example table covering transcripts, search, channels, playlists, plus documented shell differences and edge cases (live streams, language). Deductions: most of the 12 skills are narrow variants or aliases whose trigger precision cannot be checked without skill definition files; capability boundaries such as region or membership restrictions are not fully covered.
Evidence shows: a well-organized README (install, examples, pricing, FAQ, troubleshooting, license, disclosure), a complete MIT LICENSE file, install notes down to manual git clone and per-skill installs, and concrete FAQ/known limitations. Deductions: no version number, no CHANGELOG, no maintainer or update-commitment file; the relationship between the 'TranscriptAPI' copyright holder and the ZeroPointRepo owner is unexplained; the 12-skill naming includes many aliases, raising stability concerns.
Evidence shows: a clear value proposition versus yt-dlp/browser approaches, transparent free tier and pricing, and usability conveyed via example prompts. Deductions: actual output shape (JSON structure, fields) cannot be verified from the provided files; bulk channel-wide transcript fetching could consume large credit volumes while the README only says 'most operations cost 1 credit', leaving the cost envelope incomplete.
Evidence shows: SECURITY.md, a CI scanning workflow (HOL Plugin Scanner with a pinned action hash), and explicit third-party attribution/non-affiliation statements. Deductions: figures like '15M+ transcripts/month, 99.9% uptime' are untraceable; runtime support lists and the skills.sh badge cannot be cross-checked in the static file set; marketing and factual statements are interleaved in the README without separating fact from inference.
- Not found in source: rollback or recovery pathBack up first, or work on a git branch or snapshot, so its changes can be undone.
- The automated setup flow has the agent register a third-party account and handle an email OTP; prefer manual registration and confirm each step before granting access.
- The API key is written to multiple shell config files; verify where it lands and restrict visibility (mode 600).
- The skills depend entirely on an unverified paid third-party service (TranscriptAPI); independently evaluate its data policy, availability, and credit costs, which can be substantial for bulk transcript fetching.
- This is a static review with no execution; output format and endpoint behavior were not verified. Read the full skill files before installing.