---

name: cli-and-external-integration description: "Operate and extend yuxi-cli plus the external API/SSE integration flows it drives." metadata: disco-role: operating disable-model-invocation: true license: MIT

cli-and-external-integration

This sub-skill owns the public CLI surface and the external HTTP/SSE flows that the CLI exercises.

Load this sub-skill when

• The task touches yuxi, python -m yuxi_cli, or the CLI package help/config surface.
• You need remote discovery, login, whoami, status, logout, or the local browser chat helper.
• You need kb upload, kb list/files/query/open/find, or agent eval.
• You need to debug API-key auth, SSE parsing, or Langfuse dataset experiment wiring.

Do not use this sub-skill when

• The task is backend runtime behavior, deployment bootstrap, OCR/parser internals, or knowledge-base storage/indexing internals.
• You are about to write or delete remote data without explicit approval and the required credentials/services.

Owned surface

• CLI install/config/remotes/login/status/logout/chat/agent/kb commands.
• Browser chat helper and its local NDJSON bridge.
• External API auth, discovery, SSE streaming, and Langfuse experiment calls.
• Safe offline smoke checks before any live integration work.

Safe operating rules

• Run scripts/check-cli.sh first. It is offline by default.
• Treat kb upload and agent eval as side-effectful.
• Use a live remote only after discovery/version/capability checks succeed.
• Keep API keys, dataset secrets, and Langfuse credentials out of logs and examples.
• Use HTTPS for non-local remotes.

First-line entry points

• Offline smoke: scripts/check-cli.sh
• CLI entry: yuxi or python -m yuxi_cli
• Command map and proof points: references/cli-command-map.md
• External API and eval examples: references/external-api-and-eval.md
• Failure triage: references/troubleshooting.md

Change order

1. Decide whether the task is offline CLI work or a live remote flow.
2. If offline, adjust command wiring, client wrappers, or help text and keep the smoke script green.
3. If live, verify discovery, auth, and capability gates before calling the remote.
4. Match each behavior change with the CLI tests named in the command map.