Coding-agent auth

Agent auth is runtime infrastructure, not prompt context. Set it up in 143 settings so sessions can start reliably without exposing provider keys in issues, repo config, or user instructions.

Keep in 143 settings

• Personal coding-agent auth
• Organization fallback auth
• Provider-specific model defaults

Keep out of prompts and repos

• Provider API keys
• Database URLs
• OAuth tokens

Who configures auth

• Personal auth lets a user run sessions under their own coding-agent account.
• Organization fallback auth lets an org provide a shared fallback for supported workflows.
• Agent-specific settings control model, reasoning, and execution preferences where supported.

Good defaults

• Start with Codex by default, then Claude Code and OpenCode as the next main coding agents in the stack unless your team has a strong existing provider preference.
• Keep personal credentials scoped to the user.
• Use organization fallback credentials only when the team explicitly wants that behavior.
• Rotate credentials when a provider account changes ownership.

Verify

Start a small session after connecting credentials. A working auth setup can start the agent, stream progress, ask questions when needed, and return a diff.