Guide · Agent skill
How to use the KrillinAI CLI routing agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The krillinai-cli skill is an instruction package that teaches an AI agent how to build the KrillinAI command-line tool from a source checkout, choose the right command for a task, and read its JSON output. This guide explains what an agent skill is, how to install this one, how to ask for work in plain language, and what safety habits to follow. It is written for readers who have not used agent skills before.
KrillinAI CLI Routingkrillinai/OpenCreator · Apache-2.0 Download the skill Quick start
- 1Download and unzip the skill. Download the krillinai-cli ZIP and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder, or upload it under Settings → Capabilities → Skills in the Claude apps.
- 2Install Node.js and pnpm. Install the tools needed to run the repository build script.
- 3Build the KrillinAI binaries. From the repository root, run pnpm krillinai:build and confirm the CLI file exists for your platform.
- 4Create config/config.toml. Copy config-example.toml to config.toml inside config/ without overwriting an existing file, then add your provider settings.
- 5Ask for a task in plain language. Describe the output you want, such as translated subtitles or a vertical render, and name the video or text.
- 6Validate with a dry run. Ask the agent to run the command with --dry-run before any external call.
- 7Check the manifest and outputs. After a real run, review krillinai_manifest.json and the output files in your dedicated work folder.
What an agent skill is and what this one does
An agent skill is a folder with a SKILL.md file that tells an AI agent when and how to perform a kind of task. The agent reads the skill when your request matches its description, then follows its instructions and supporting reference files.
This skill is a top-level routing skill for KrillinAI command-line work. It helps the agent map your intent to a supported command: subtitles, dubbing, landscape or vertical renders, cover images, speech, or voice lists. It also tells the agent how to handle outputs and errors.
- Generates source, target, bilingual, and short vertical subtitles through the subtitle command
- Generates target-language dubbing from subtitles through the tts command
- Creates landscape videos with render-horizontal and portrait videos with render-vertical
- Creates a cover from a complete image prompt, generates speech from text, and lists voice codes for Aliyun, OpenAI, or MiniMax
Install the krillinai-cli skill
Download the skill ZIP from the page where you found this guide, then choose one of the install routes below.
For Claude Code, unzip the archive into ~/.claude/skills/ for a personal install, or into a project's .claude/skills/ folder to share it with one repository. For the Claude apps, open Settings, go to Capabilities, and upload the skill under Skills.
- Personal: unzip into ~/.claude/skills/
- Project: unzip into .claude/skills/ inside your project
- Claude apps: Settings → Capabilities → Skills, then upload the ZIP
Prepare the KrillinAI CLI before asking for work
The skill only routes commands. It cannot run KrillinAI by itself, so you need the repository checkout or an unpacked release archive with a built binary. Build both binaries through the repository script, then confirm the CLI exists for your platform.
The CLI reads config/config.toml relative to its working directory. Create that file from config/config-example.toml without overwriting an existing one, and fill in your provider settings. Real runs from a source checkout use runtime/krillinai as the working directory. A release archive should run from its unpacked package root.
- Node.js and pnpm, to run the repository build script
- A built krillinai-cli binary from the checkout or an unpacked release archive
- config/config.toml created from config-example.toml, holding provider settings and likely API keys
pnpm krillinai:buildTrigger the skill with natural requests
You do not need to name a command. Describe the result you want and mention the video or text involved. The agent should pick the matching command from the routing table and check the output folder before running anything.
Be specific about languages, the video source, and the format. For example, say which origin and target languages you need, and whether you want a landscape or a vertical result.
- "Make translated English-to-Chinese subtitles for this YouTube video and use a dedicated task folder."
- "Render the subtitled project as a vertical short in my demo workdir."
- "List the OpenAI voice codes I can use for dubbing."
- "Check that this cover command is valid with a dry run before generating anything."
Follow a minimal subtitle-then-render workflow
The skill's example runs a subtitle stage on a video URL with an origin language, a target language, a work folder, caption source settings, and video preparation, then runs a vertical render on the same work folder. The second stage reuses the outputs recorded for the first, so the agent does not need to guess filenames.
Replace VIDEO_ID with your own video identifier and keep the commands running from runtime/krillinai, as the skill directs.
- Run subtitle first, then render-vertical on the same --workdir
- Check krillinai_manifest.json and the output files after each real stage
(cd "$REPO_ROOT/runtime/krillinai" && "$KRILLINAI_CLI" subtitle "https://www.youtube.com/watch?v=VIDEO_ID" --origin-lang en --target-lang zh_cn --workdir "$WORKDIR" --caption-source any --prepare-video)Understand safety, scripts and network use
This skill runs scripts and uses the network. The CLI runs locally but makes external provider calls, so the commands can send requests to outside services using the keys in your config file. Treat untrusted video URLs and unknown configuration files with caution.
The skill's habits reduce risk: use a dedicated --workdir instead of the repo root, validate command shape with --dry-run before external calls, and avoid rerunning expensive stages when the manifest already has valid outputs.
- Use a dedicated --workdir so outputs do not mix with the repo root
- Run --dry-run first to check command shape before any external call
- Do not run untrusted video URLs or unknown config files
- Keep API keys in config/config.toml and out of shared files
Know the limits and error handling
The pipeline command only validates plans in dry-run mode. Non-dry-run pipeline execution returns unsupported_command, and status is reserved and unsupported. The cover command accepts a complete text prompt and size only, so it cannot use a reference image. The voices command supports aliyun, openai, and minimax, not an Edge TTS catalog.
When a command fails, the agent should classify the error by its kind: usage, retryable, dependency, or internal. Parsing stdout as JSON lines matters because the final response is the object containing ok, and progress frames may come first.
- pipeline: dry-run validation only
- status: reserved and unsupported
- cover: text prompt and size only
- voices: aliyun, openai, minimax
What people use it for
- Generate translated subtitles, including bilingual and short vertical versions, for a video
- Produce target-language dubbing from existing subtitles
- Create landscape or portrait short-form renders from a prepared work folder
- Generate a cover image from a complete prompt or speech from plain text
- List provider voice codes before choosing a dubbing voice
Frequently asked questions
What is the krillinai-cli skill?
It is an agent skill that teaches an AI agent how to build the KrillinAI CLI, choose the right command for a task, and read its JSON, manifest and output files. It routes requests; it does not run KrillinAI on its own.
What do I need before installing it?
You need the KrillinAI repository or an unpacked release archive, a built krillinai-cli binary, Node.js and pnpm for the build script, and a config/config.toml file created from config-example.toml.
Where do I install the skill?
Unzip the ZIP into ~/.claude/skills/ for a personal install or into a project's .claude/skills/ folder. In the Claude apps, upload it under Settings → Capabilities → Skills.
Is the skill free and what is its licence?
The skill is published under the Apache-2.0 licence. This page does not state pricing, so check the repository and any provider accounts you use for costs.
Why does pipeline return unsupported_command?
The pipeline command only validates plans in dry-run mode. Non-dry-run execution returns unsupported_command, so use pipeline with --dry-run to check a plan.
Does the skill run safely on untrusted videos?
Use caution. The skill runs scripts and uses the network, and the CLI makes external provider calls with keys from your config. Use a dedicated work folder and dry runs, and avoid untrusted URLs or unknown config files.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use this skill if you already work in the KrillinAI repository and want an agent to drive its CLI for subtitles, dubbing, or short-form renders. Skip it if you do not have the repository or a built binary, or if you need the pipeline command to run multi-stage work, since that is not supported yet.
Download the skill