Coding agents

Prepare

Install the CLI first. Schema inspection and Skills generation need no key. Configure account authentication yourself before allowing real paid operations. Do not paste your key into the coding assistant.

kitta skills generate --api-url https://fishaudio.org --output-dir .agents/skills/kitta-audio
kitta audio tts --api-url https://fishaudio.org --schema

Review the generated Skills files and enable them according to your coding assistant's instructions. Writing the directory does not prove the assistant has loaded the skill. The generated files use generic CLI defaults; skills generate does not save your site origin. Give the assistant the site instructions below alongside the generated skill.

Inspect before execution

kitta audio tts --api-url https://fishaudio.org --voice VOICE_ID --text "Hello from Kitta" --dry-run

Replace the voice ID. Dry-run locally validates and displays the request without network access or submission records. Authorization is redacted, but input text appears in the output. It does not check live model availability, balance or prices.

Suggested assistant instructions

For every CLI call, explicitly pass --api-url https://fishaudio.org for authentication, generation, task queries and recovery. Use only the API key and task IDs belonging to this origin. The generated skill does not bind the site, and the CLI default remains https://kittaai.com.

Before paid operations, confirm the text, voice, output file and scope. Do not read or display secrets. Retain a business idempotency key for new TTS operations and reuse the original key and input for recovery. Report task IDs on failure; do not automatically resubmit cloning, design or transcription.

The CLI provides terminal commands and Skills files; do not describe it as a deployed MCP service.

Failure and next steps

Fix invalid inputs before making a real submission. If the assistant does not know the remote outcome, inspect the task first. Recovery guide · Automation