Gate your GitHub token from an AI agent

When: You use Cursor / Claude Code / Codex and also use gh. You want the agent constrained, while your own terminal stays productive.

Outcome: Terminal can do normal gh work (Trusted default). AI harness defaults to Read (list/view style); writes and secret-reveal need a prompt or a higher policy level.

# 1. Install the latest release (other ways: the Install page)
powershell -NoProfile -ExecutionPolicy Bypass -File .\scripts\Install-CmdWarden.ps1 -Desktop

# 2. Health
cw doctor

# 3. Enroll YOUR normal shell (Windows Terminal / PowerShell - not inside the agent)
cw whoami
cw policy enroll --kind terminal

# 4. Enroll the AI harness (run these from that harness integrated terminal)
cw whoami
cw policy enroll --kind ai-harness

# 5. Harden gh (pin + PATH shim + import token into Vault as GH_TOKEN)
cw harden gh

# 6. New shell so PATH sees the shim, then verify
where.exe gh
gh auth status
cw policy list

Expect: cw doctor shows Session Agent UP; enrollments appear in policy list; first gh on PATH is under CmdWarden shims; agent-driven gh that needs more than Read may prompt or block depending on command class.

Why two enrollments? Policy is tool + launcher. Same gh, different caller (you vs agent) can get different levels.