Renoise CLI — terminals, scripts, and local helpers

Platform URL: https://www.renoise.ai — never renoise.com.

One interface in conversations

In a conversation, all Renoise work goes through the Renoise MCP tools this plugin connects: list_models, estimate_cost, create_tasks, wait_tasks, show_result, upload_files, download_results, analyze_media, search_materials, cancel_tasks, save_result_as_material, get_account. They are the same tools on every host, and their instructions carry the approval rule for paid work.

Never create a paid task from a conversation through renoise task create, batch-generate.sh, or any other CLI path: that would bypass the approval rule. If the Renoise tools are not available, read ../renoise-setup/SKILL.md and connect them; do not fall back to the CLI.

The native renoise binary remains for what the user runs themselves: shell scripts, batch jobs, and CI. Point them to renoise --help and renoise <command> --help rather than documenting flags here.

Install or update the CLI

Needed only for the annotation board, the helpers below, or the user's own scripts:

node "${CLAUDE_SKILL_DIR}/../renoise-setup/scripts/install-cli.mjs" --ensure
export PATH="${HOME}/.local/bin:${PATH}"
renoise version
renoise auth status --json

On Windows PowerShell, run the same node ... install-cli.mjs --ensure, then prepend $env:LOCALAPPDATA\Renoise\bin to $env:Path. If --ensure fails or authentication is missing, follow ../renoise-setup/SKILL.md. Never ask the user to paste an API key into chat.

Local production helpers

These run locally and never spend credits:

# Preview/QC and grid helpers used by director workflows
bash ${CLAUDE_SKILL_DIR}/scripts/qc-preview.sh --videos-dir <videos-dir>
bash ${CLAUDE_SKILL_DIR}/scripts/split-grid.sh storyboard.png output-dir/ 2 3
node ${CLAUDE_SKILL_DIR}/scripts/match-materials.mjs --pool material-pool.json --shots project.json
python3 ${CLAUDE_SKILL_DIR}/scripts/analyze-beats.py <audio-file>
node ${CLAUDE_SKILL_DIR}/scripts/generate-preview.mjs <project-dir>   # HTML storyboard preview

material-ingest.mjs (upload and analyze a folder) and batch-generate.sh (paid batch generation) call the CLI with the user's credential: they are terminal tools for the user to run. In a conversation, upload with upload_files, analyze with analyze_media, and generate with create_tasks. Inspect a helper's help/source before use; never run them in a Hosted Agent.

Errors

Use the CLI's JSON error and exit code; do not infer API behavior from copied endpoint documentation. JSON errors expose retryable and, after task creation, taskId; retry only the wait/query operation for that ID, never the paid create operation.

Do not pre-screen prompts or materials and do not infer a moderation result yourself. Continue normally unless the CLI/API returns an explicit content-review error (INPUT_* / OUTPUT_*). On that actual error, report it and stop; never retry the paid create operation or suggest bypasses.