[dines-oai]

What and why

Adds CodexAxonConnection.readConfigRequirements(), a wrapper for the Codex app-server's configRequirements/read. It also exports the ConfigRequirements and ConfigRequirementsReadResponse types.

Codex sessions signed in to a ChatGPT Business, Enterprise, or Edu workspace inherit the workspace admin's managed policy, for example allowed sandbox modes or permission profiles. If a client asks for something the policy forbids, thread/start can fail, or Codex can fall back to its defaults. Today a client cannot ask for the policy first, because request() is private and readConfig() (config/read) does not include requirements. Reflex needs this to choose its Codex agent permissions up front (runloopai/reflex#5341).

Change

• readConfigRequirements() sends configRequirements/read with the params field omitted. The wire type is Option<()>, matching getAccountRateLimits().
• The response's requirements is null when no policy applies.
• I added it to the wire-params test and to the method tables in sdk/README.md and sdk/AGENTS.md.

Checklist

• PR title follows <type>(<scope>): <description> format
• bun run check passes (lint + format)
• bun run build passes
• bun run test passes (653 tests)
• SDK documentation updated

I checked the method by hand against the Codex 0.159.1 app-server: it returns {"requirements": null} without a policy, and the allowed sandbox modes or profiles when /etc/codex/requirements.toml sets them.

🤖 Generated with Claude Code

[reflex-loop]

⚠️ Reflex agent status: Error

Agent	Status
agt_6yKSJYRvCGSmT7bxGP3H4T	⚠️ Error

This PR was created by Reflex.

_{This comment updates in place as agents work.}