Agents

Live checks always run through an installed agent CLI (Cursor Agent, Codex, or Claude Code). There is no built-in default agent. Checks use your agent subscription and its quota.

Sign in

After the agent CLI is on PATH, sign in with Skill Gate. That starts the agent's official login in your terminal:

shell
skillgate login cursor
skillgate login claude
skillgate login codex
skillgate login --status

skillgate login with no name signs in installed agents that still need a session. skillgate doctor prints the same installed / logged in / none table.

If you are already signed in, skillgate login cursor still starts Cursor's official browser login. Leave that window open until it finishes. Do not interrupt or close it mid-flow: that can leave the session half-done. Confirm with skillgate login --status.

Shell keys override login

A CURSOR_API_KEY (or OPENAI_API_KEY / ANTHROPIC_API_KEY) set in your shell overrides the subscription login. Unset it to use your plan:

shell
unset CURSOR_API_KEY

Cursor Agent CLI

Codex CLI

Claude Code

Auto selection

Default harness is auto. Skill Gate picks the first agent that is installed and signed in, in this order:

  1. Cursor Agent CLI
  2. Codex CLI
  3. Claude Code

Set a default with SKILLGATE_HARNESS or override per run with --harness on check, run, and compare arms.

shell
skillgate check skills/pdf --harness claude
export SKILLGATE_HARNESS=cursor-agent
skillgate check skills/pdf

Exit 2: no agent

When auto cannot find a ready agent (or you pick a harness that is not signed in), the CLI prints:

shell
No agent is available to run checks.

Install the CLI, run skillgate login for that agent, then retry. The process exits with code 2.

Quota

Each check is a live run of your signed-in agent. It uses that subscription and counts against its quota, the same as using the agent yourself.