LinkAI CLI

LinkAI CLI (agent edition)

This skill is optimized for agent invocation, not interactive human use. Default to JSON, non-streaming, non-interactive flags.

If linkai is not installed, see the fallback install note at the end of this file.

Agent defaults — always apply

Capability map

Decision flow

1. Pick the command from the capability map.
2. Open the matching reference for required flags and JSON output fields.
3. Resolve missing IDs (app_code, kb_code, db_code, plugin_code) by listing first with --json. Don't guess codes.
4. Run with --json. Parse the result; surface only what the user needs.
5. On non-zero exit, classify via errors.md. For scope errors, stop and ask the user to re-login — do not retry.

Pre-flight

Before the first LinkAI call in a session, verify auth: linkai auth status --json. If status is not valid, run the login flow below.

Login (agent flow)

Login needs the user to authorize in a browser. After you send the link, do NOT stop to wait for the user to reply — make the poll your next tool call.

1. Get the URL and tell the user to open it and authorize:

   linkai auth login --no-wait --json

2. On your next tool call, poll until they finish — each call blocks up to --wait seconds:

   linkai auth login --device-code <code> --wait 60 --json

   Re-run the same command while event is authorization_pending (the poll blocks, so looping is not busy-spinning); stop on authorization_complete (then continue the user's original request) or authorization_failed.

See auth.md for the JSON fields and event handling.

Install (fallback) {#install-fallback}

If the linkai command is missing, install it (no interaction needed):

npm i -g linkai-cli   # when Node.js is available

If npm is unavailable or fails, see install.md for the
install-script and manual-download methods (macOS/Linux/Windows). Verify with
linkai --version.