Setup guidance specific to automated agents (Claude Code, Cursor, etc.) working in this repo. The full human-oriented setup — toolchain, Cursor IDE, and the step-by-step for internal resource access — lives in Setup.md; this page is only what differs for an agent, so you don't have to read the interactive human steps to know how to behave.
These steps require a human at the keyboard. Hand them off — don't attempt them:
| Step | Why it's human-only |
|---|---|
Browser-based Okta SSO (generate-aws-config.sh, awslogin) |
Needs an interactive browser + MFA |
Supplying Mongo credentials for SC_MONGODB_URI |
Live in 1Password — never touch 1Password or the op CLI |
Opening a production tunnel (lifi-connect prod …) |
Persistent prod channel, gated behind explicit human approval |
Installing CLIs, running generate-kubeconfig.sh, and editing repo files are
fine — non-interactive and secret-free. The commands themselves are in
Setup.md → Accessing LI.FI resources;
this page doesn't repeat them.
- Shell aliases don't exist. Use
aws sso login --sso-session LIFI, not theawsloginalias. - PATH is not the login shell's. Prepend tool locations explicitly:
export PATH="$HOME/.foundry/bin:$HOME/.bun/bin:$HOME/.local/bin:$PATH"(~/.local/binis wherelifi-connectinstalls). - Assume the human has already authenticated and started any needed tunnel. Don't run the browser login or open the prod tunnel yourself — and don't ask the human to confirm it's up before you start. Proceed optimistically: run the tunnel-dependent command, and only surface a problem if it actually fails (Safe/Mongo scripts fail fast with an actionable error — see below).
getSafeMongoCollection() (in script/deploy/safe/safe-utils.ts) throws a
fail-fast, actionable error naming lifi-connect prod smart-contracts when the
tunnel is down. Surface that to the human rather than trying to fix access
yourself. Never read .env secret values, and never write a resolved
SC_MONGODB_URI back into the repo.