Live-Stream Clips

Before you start: read this whole install prompt, tell me in a few lines what this skill does and what installing it involves, and wait for my go-ahead before you do anything else. Prerequisites: none for the install itself: no account, no API key, no paid service, and nothing else to install first. For using the skill afterwards, read this first: its only bundled script, scripts/live_clip.py, is local-only and needs Python 3 (the script imports only standard-library modules: argparse, csv, os, re, subprocess, sys) plus the ffmpeg and ffprobe command-line tools on PATH (the script calls them locally to probe the video duration, detect pauses and render the clips; the rendering uses ffmpeg's libx264, loudnorm, blur-fill, drawtext and subtitles filters). The title and subtitle rendering asks ffmpeg for the font "Noto Sans CJK SC" by default (hard-coded in the script, no --font option and no font check in the code), so a matching CJK font needs to be available to ffmpeg if I want titles or burned subtitles to render as intended. Pillow, NumPy and edge-tts are NOT used by this skill: the repository root requirements.txt lists pillow>=10, numpy>=1.24 and edge-tts>=6.1, but live_clip.py imports none of them — they belong to other skills in the repository. No generation API, no text-to-speech service, no stream or VOD download, no key and no online link are used by this skill at all: the livestream recording is a local file I supply. Install "live-stream-clips" for me. It turns a long ecommerce livestream recording into a batch of 15–60 second vertical clips: choose the segments three ways — (A) timestamps I or the agent pick, listed in a cuts.csv file (start,end,title), (B) keyword hits in an SRT transcript I supply (each hit takes a --window-second segment, 30% before and 70% after the hit, overlapping windows are merged), or (C) automatic proposals split at pauses detected in the audio when there is no transcript (--propose, which writes a candidate CSV for me or the agent to pick from) — then export every chosen clip as a 9:16 MP4 (landscape sources get a blurred-fill background with the picture centred and complete), with an optional title burned on top, optional burned subtitles (--burn-subs), and loudness normalised. SKILL.md also gives highlight signals (live demo, Q&A answered, emotion peak, rules explained), a transcript prompt template for picking 8 clips of 20–40 seconds (CSV start,end,title, title at most 14 characters, no absolute claims, no specific price promises), rules that every clip must make sense on its own with a hook in the first 3 seconds and cuts on sentence boundaries, and a checklist including that any price/offer/stock information still shown must still be valid and that I must own the livestream or have the streamer/brand's permission to repost it. Repository: https://github.com/xianyu110/ecommerce-video-skills (the skill is the whole skills/live-stream-clips/ folder at the pinned commit: SKILL.md plus its scripts/ folder, 2 files) Pinned version: install the code at commit 96c49b349b7d975bda8f05ce6669c1d33500c269 exactly. Do not substitute a newer version. If this commit is unreachable, stop and tell me. 1. Clone https://github.com/xianyu110/ecommerce-video-skills into a temporary folder, check out commit 96c49b349b7d975bda8f05ce6669c1d33500c269, and copy the skills/live-stream-clips/ folder from that checkout. Then compare every file in the folder with the fingerprints recorded for this listing on 2026-10-19: [FINGERPRINTS] - SKILL.md: SHA-256 8b88c53187262245f13ad21520fedbb7d32bd40656caf4efa8f50e68d16803fc - scripts/live_clip.py: SHA-256 f7f83f215647fe772e9bf32024114676ddd32d18c76abec67209327991cc5eb3 If a file does not match, it changed since the listing was prepared: do not install it yet, tell me what you found and let me decide. 2. Read SKILL.md in full and read the only script in scripts/ (live_clip.py), and tell me in plain words: what the skill does when used, which services it connects to, whether it reads or sends any personal information or keys, and anything that seems unusual. Stop on any red flag and tell me before going further. Confirm these points from the files: - The skill folder holds 2 files only: SKILL.md (3,851 bytes) and one Python script in scripts/ (live_clip.py, 5,544 bytes). That script is a byte-identical copy of the same-named script at the repository root (scripts/live_clip.py, same SHA-256): the skill is self-contained in its own folder and needs no file from elsewhere in the repository — not the other 12 skills, not the root scripts/, not the README, and not the repository requirements.txt at install time. SKILL.md only ever calls the script sitting next to it (scripts/live_clip.py, relative to the skill folder). Step 5 of SKILL.md names two sibling skills in words only (use the hook types from hook-3s-script to write 2 titles per clip, then hand them to cover-title-ab): that is a workflow suggestion in the text, no file from either sibling is read or called by live_clip.py, and neither sibling is required to install or to run this skill's script. - Nothing is sent anywhere during installation itself; the only download at install time is the skill folder from GitHub at the pinned commit. No script is run during installation, no key is asked for, and there is no telemetry in the folder. - What leaves this machine at use time, disclosed here before it ever happens: nothing. Using the skill is a separate step that only starts when I ask for clips, and it sends nothing to anyone: scripts/live_clip.py makes no network call at all — it imports only argparse, csv, os, re, subprocess and sys, contains no URL, no API client, no download code and no key handling, and SKILL.md defines no API-key route, no online-link route, no text-to-speech route and no stream/VOD download route for this skill. The recording is never fetched by the skill: I supply it as a local file (mp4/flv/mov, landscape or portrait). Three things in SKILL.md are not sends by the skill: (1) the optional SRT transcript is a local file I supply — SKILL.md says it can come from a platform export or from any ASR tool I choose myself, and obtaining it that way is my own separate step outside this skill, before I ever run it, not something the skill does; (2) the transcript prompt template in SKILL.md is text I use with you locally in our conversation — if I paste a transcript to ask for clip picks, that text stays in our conversation as my input to you, it is not sent by the script to any service; (3) publishing the finished clips to Douyin / WeChat Channels / TikTok or anywhere else happens manually, by me — the skill itself never contacts any platform. - The local-only parts at use time: scripts/live_clip.py runs entirely on this machine — it calls the local ffprobe to get the video duration; with --propose it calls the local ffmpeg silencedetect filter (noise=-35dB, d=0.6) to propose segments split at pauses and prints a CSV to standard output; with --cuts it reads my local cuts.csv; with --srt and --keywords it reads my local SRT and finds the keyword hits locally; then for each segment it calls the local ffmpeg to export a 9:16 clip (default 1080x1920, --size to change it) with a blurred-fill background for landscape sources, the optional title drawn on top and the optional subtitles burned in, encoded as libx264/AAC with loudness normalised (loudnorm I=-14:TP=-1.5:LRA=11), and writes clip_01.mp4, clip_02.mp4 and so on to the local output folder (default out/clips). The highlight signals, the prompt template and the checklist in SKILL.md also stay local, between you and me. Rules from SKILL.md: only clip a livestream I own or have the streamer/brand's permission to repost, every clip must stand on its own with a hook in the first 3 seconds, and any price, offer or stock information that is no longer valid must be cut out or the clip taken down. - Do not install ffmpeg/ffprobe or anything else during this install, do not run live_clip.py, and do not generate any clip: setup of the use-time dependencies is my decision, later, when I actually ask for clips. 3. When I say "yes, install", install it as a skill: copy the whole live-stream-clips folder into the agent's skills directory, in a folder named "live-stream-clips", keeping all 2 files (SKILL.md at its root and the one script in its scripts/ subfolder). A manual clone followed by copying the skills/live-stream-clips folder into ~/.claude/skills/ is the route described here; any other route is fine only as long as every installed file matches its fingerprint. 4. Verify with no network calls: frontmatter valid (name: live-stream-clips), all 2 files in place with matching fingerprints. Do not run live_clip.py to verify. 5. Report what was installed, where, and what I still need to do myself (for example: provide a livestream recording I own or have permission to repost and, if I have one, an SRT transcript, plus the clip length, number of clips and the product or topic to highlight; have Python 3 and ffmpeg/ffprobe on PATH, and a CJK font available to ffmpeg, if I want the script to render clips; using the skill is a separate step that only starts when I ask for it, and no use-time send happens on it at all, as disclosed above). GitHub is optional: if I have a GitHub account or the gh CLI, you may use it; otherwise public access is fine. Never require it unless it's in the prerequisites above. Rules: don't touch anything outside the temp folder and the install target. If anything looks off, stop and ask me.