Skip to content
/
Try

Using Skink inside EveryAI

There is nothing to configure. When EveryAI starts Skink in its terminal, it sets two environment variables, and Skink hands language servers and AI work over to EveryAI. The status bar then shows EveryAI.

Without those variables, Skink runs standalone and never contacts anything.

  • Language servers come from EveryAI. Completion, hover, go to definition, references, rename, format and diagnostics work the same, but you do not need the servers installed for Skink.
  • The AI prompt box (Ctrl+G) and the agent panel (Alt+Shift+A) run on EveryAI’s AI instead of a local ACP agent.
  • Models: Tab in the prompt box and /model pick from EveryAI’s model list. ai.model and the per-mode ai.model.MODE settings match EveryAI’s model ids or names.
  • Font size: :fontsize +1, -1 and 0 ask EveryAI to change the terminal font. See Zoom and text size.

Everything else (editing, tabs, terminal tabs, git, tasks, settings) is the same as standalone.

Variable Meaning
SKINK_HOST_URL EveryAI’s MCP endpoint on loopback, for example http://127.0.0.1:41234/mcp
SKINK_HOST_TOKEN The bearer token for this session

Both must be set and non-empty.

Skink talks MCP over HTTP to 127.0.0.1 only, and checks the connection with short timeouts at start-up.

Running inside EveryAI never makes Skink worse than running alone:

  • If the host does not answer at start-up, Skink stays standalone silently.
  • If a call fails mid-session, that feature falls back to Skink’s own implementation (local language servers, local ACP agent), and the status bar notes that the host is unavailable. Skink tries the host again on the next request.
  • You can always cancel a running AI request with Esc.