concinno-skills-video 0.1.0

YouTube + video workflow agent skills for Concinno — yt-dlp (Unlicense), youtube-transcript-api (MIT), moviepy (MIT). Requires system ffmpeg.

concinno-skills-video

YouTube + video workflow agent skills for Concinno.

Status

MVP (0.1.0) — three tools covering the public-video workflow end-to-end:

Tool	Library	Purpose
YouTubeDownload	yt-dlp (Unlicense)	download video / extract audio / fetch metadata
YouTubeTranscript	youtube-transcript-api (MIT)	fetch YouTube-provided captions (no STT)
VideoEdit	moviepy (MIT)	subclip / concat / thumbnail

Private / playlist / upload support is deliberately not in 0.1.0 — those paths need OAuth and will land in a later minor release. The _auth.py placeholder exists so 0.2.0 can add them without reshuffling the package layout.

Install

pip install concinno-skills-video

You must also install ffmpeg — the package does not bundle it.

• macOS: brew install ffmpeg
• Debian / Ubuntu: sudo apt install ffmpeg
• Windows: choco install ffmpeg or download from https://ffmpeg.org/

Each tool checks shutil.which("ffmpeg") on first use and returns a structured error if it's missing rather than crashing.

Legal & operational notes

• yt-dlp is Unlicense — you can embed it freely, but downloading YouTube content may still violate YouTube's Terms of Service and your local copyright law. This library does not grant you permission to download content you don't have rights to. DMCA takedowns, IP bans, and content-policy enforcement are entirely downstream of this code.
• youtube-transcript-api hits YouTube's public caption endpoints; sustained bursty use can trigger IP throttling or short bans. The tool is marked is_concurrency_safe = False for this reason.
• Size cap: YouTubeDownload refuses any request whose expected filesize exceeds 500 MB unless you pass force=True. This is a guardrail to prevent agent-triggered accidental bandwidth burn, not a security boundary.
• Path containment: all file paths (output_dir, input_path, output_path) must resolve under Path.home() or Path.cwd() — arbitrary absolute paths are rejected (Concinno BoundaryGuard rule).

Usage via Concinno ToolRegistry

When the consumer sets CONCINNO_LOAD_PLUGINS=1, the default registry auto-mounts all three tools:

import os
os.environ["CONCINNO_LOAD_PLUGINS"] = "1"

from concinno.tools.registry import get_default_registry

reg = get_default_registry()
assert {"YouTubeDownload", "YouTubeTranscript", "VideoEdit"} <= set(reg.list_deferred())

tool = reg.get("YouTubeTranscript")
tool.call(action="get", video_id="dQw4w9WgXcQ", lang="en")

Direct Python usage

from concinno_skills_video import (
    YouTubeDownload, YouTubeTranscript, VideoEdit,
)

# Metadata only (no download)
info = YouTubeDownload().call(
    action="info",
    url="https://youtu.be/dQw4w9WgXcQ",
)

# Audio extract
YouTubeDownload().call(
    action="audio",
    url="https://youtu.be/dQw4w9WgXcQ",
    output_dir="~/Downloads",
)

# Transcript
segs = YouTubeTranscript().call(
    action="get",
    video_id="dQw4w9WgXcQ",
    lang="en",
)

# Edit: cut a segment
VideoEdit().call(
    action="subclip",
    input_path="~/Downloads/input.mp4",
    output_path="~/Downloads/cut.mp4",
    start=10.0,
    end=20.0,
)

# Edit: concat
VideoEdit().call(
    action="concat",
    input_paths=["~/Downloads/a.mp4", "~/Downloads/b.mp4"],
    output_path="~/Downloads/merged.mp4",
)

# Edit: extract a frame as PNG/JPEG
VideoEdit().call(
    action="thumbnail",
    input_path="~/Downloads/input.mp4",
    output_path="~/Downloads/thumb.png",
    time=5.0,
)

All methods return either {"ok": True, ...} on success or {"error": "..."} on failure — same shape as the other Concinno built-in tools.

License

Apache-2.0. See LICENSE in the Concinno monorepo.