Custom Agents MCP, sandbox restriction definitions broken
Nobody has claimed this yet.
- Dominant language
- Rust
- Stars
- 125k
- Forks
- 19.4k
- PR merge metrics
- PR metrics pending
Description
What version of the Codex App are you using (From “About Codex” dialog)?
26.818.41509
What subscription do you have?
Pro 5x
What platform is your computer?
No response
What issue are you seeing?
https://github.com/openai/codex/pull/39299
This breaks subagents v1 and v2. This also directly contradicts public documentation https://learn.chatgpt.com/docs/agent-configuration/subagents?surface=app#app-custom-agents.
I can also imagine this increases context usage and token burn.
What steps can reproduce the bug?
Define subagents, sandbox_mode, anything. It's all broken.
What is the expected behavior?
I can restrict readonly agents to be readonly. I can give my researcher agents context7 without giving all my agents access. I can not give Luna access to AWS.
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start by reviewing pull request #39299 and the linked custom-agents documentation, then reproduce the failure with subagents, sandbox_mode, and MCP restrictions. Done means readonly agents remain readonly and per-agent access to tools such as Context7 or AWS can be restricted as configured.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- rust
- Domain
- authorization, security
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Active
- Clarity
- Needs clarification
- Newbie friendliness
- 42/100