How to use the KrillinAI Subtitle Generator agent skill

What an agent skill is and what this one does

An agent skill is a folder containing a SKILL.md file. The file gives an AI agent instructions for one kind of task. This skill covers the subtitle stage of the KrillinAI workflow.

Given a YouTube, Bilibili, or local video, the agent first tries to download the platform's existing captions. If none exist, it falls back to speech transcription with Whisper. It can then produce translated subtitles, bilingual subtitles, and a short-form mixed subtitle file.

• Platform captions are used first when available (--caption-source any).
• Whisper transcription is used as the fallback, or forced with --caption-source whisper.
• Outputs include origin_language_srt.srt, target_language_srt.srt, bilingual_srt.srt and short_origin_mixed_srt.srt.
• The original video is kept when --prepare-video is requested, for later render stages.

What you need before installing

The skill only orchestrates the KrillinAI CLI. It cannot work alone, so the KrillinAI CLI binary must already be installed on your machine. The skill finds it through the KRILLINAI_CLI and KRILLINAI_CWD environment variables.

The CLI configuration is described in skills/krillinai-cli/references/cli-contract.md. That configuration may require API keys for AI and translation services; the skill file itself does not specify which keys. URL inputs need network access that yt-dlp can use.

• KrillinAI CLI binary, with KRILLINAI_CLI and KRILLINAI_CWD set
• The krillinai-cli reference file for configuration
• API keys for AI and translation services, if your CLI setup requires them
• Network access for YouTube, Bilibili or other yt-dlp supported links
• Whisper transcription capability, local or configured through the CLI

Install the skill from ZIP

Download the ZIP for krillinai/OpenCreator skills/krillinai-subtitle from this page and unzip it. The package contains a single file, SKILL.md.

For a personal install, place the folder in ~/.claude/skills/. For a single project, place it in that project's .claude/skills/ folder. Claude picks up the skill from these locations.

In the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.

• Personal: ~/.claude/skills/krillinai-subtitle/
• Project: .claude/skills/krillinai-subtitle/ inside your project
• Claude apps: Settings → Capabilities → Skills, then upload

Ask for subtitles in plain language

You do not need to name the skill or type a command. Describe the video and the result you want, for example a Chinese translation of an English YouTube video, with a bilingual file. The agent should match the request to this skill's description and then read the CLI contract before running anything.

Be specific about the source language, the target language, and whether you need bilingual or short-form output. If you have a local file, give its absolute path, since the skill accepts local:/abs/path/video.mp4 or a path the CLI accepts.

• "Make English subtitles for this YouTube link and translate them to Simplified Chinese as a bilingual SRT."
• "Transcribe this lecture at /abs/path/lecture.mp4 in Japanese and give me the origin-language SRT."
• "Create subtitles for this Bilibili video, and a short vertical mixed subtitle file."

The command the agent runs

The skill runs the KrillinAI subtitle command from the CLI working directory. Source and target language, the task directory, caption source, and the video-preparation flag are all set in the command. The agent substitutes your input and working directory.

• --origin-lang sets the source language, such as en, zh or ja.
• --target-lang sets the target language, such as zh_cn.
• --workdir sets a dedicated task directory.
• --caption-source any prefers platform captions; whisper forces transcription.
• --prepare-video requires the original video for later render stages.
• --bilingual-top=true puts the target language on top in bilingual SRT.
• --subtitle-style-file merges a JSON subtitle-style override.
• --dry-run validates the command without downloads or AI calls.

Check the results

A successful real run should print terminal JSON with "ok": true. The agent should then confirm that krillinai_manifest.json exists under the work directory and read output paths from it. A dry run does not write a manifest or output files, so it only proves the command shape is valid.

Spot-check the bilingual file for correct line ordering and the short mixed file for readability. If you asked for video preparation, the origin_video output should exist and be non-empty.

• Terminal JSON shows "ok": true
• krillinai_manifest.json exists in --workdir
• bilingual_srt.srt lines are in the expected order
• short_origin_mixed_srt.srt is readable
• origin_video.mp4 is non-empty when --prepare-video was requested

Safety notes: scripts and network

This skill runs scripts and uses the network. It fetches media from the internet via yt-dlp and may call external download and AI services. Only process videos you have permission to use. The skill is not suitable for a fully offline workflow.

Use a dedicated --workdir for each task so outputs and intermediate files stay together. Review the CLI configuration in the reference file before adding API keys.

• Use only videos you own or have permission to process.
• Expect network calls to download and AI services.
• Keep each task in its own work directory.