[abhinav-oai]

Summary

This PR enables prompt hooks end to end.

It defines the prompt-hook configuration shape, discovers and trust-checks configured prompt handlers, and connects the execution layer from the first PR to Codex's model client.

After this PR, a trusted type = "prompt" hook can run for supported events and affect the hook outcome.

Configuration and discovery

Prompt hooks support:

• prompt
• model
• timeout
• statusMessage
• continueOnBlock

Discovery validates the event, rejects empty prompts, applies defaults, includes the full definition in the trust hash, and constructs runnable prompt handlers.

Model execution

Each hook runs as a separate model request containing only the rendered hook prompt and hook input.

The request:

• has no conversation history
• has no tools
• does not inherit developer instructions
• uses the thread model unless the hook selects another model
• uses an isolated ModelClientSession

The isolated session is important because a normal model session reads and replaces the main conversation's cached WebSocket continuation state. A hook request must not disturb the next normal turn.

Example

[[hooks.UserPromptSubmit.hooks]]
type = "prompt"
statusMessage = "Checking whether the request is actionable"
prompt = "Evaluate if the ask is well specified: $ARGUMENTS. Block requests that are underspecified!"

Given:

› help

The hook can return:

• UserPromptSubmit hook (blocked)
  feedback: Request is underspecified: clarify what help is needed and the desired outcome.

Stack

1. Add prompt-handler execution to codex-hooks #26267
2. (this PR) Enable configured prompt hooks in Codex #24634
3. Expose prompt hooks to clients #26268

[abhinav-oai]

https://code.claude.com/docs/en/hooks#response-schema for reference

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: cd80c08856

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[abhinav-oai]

A normal ModelClientSession checks out the main session’s cached WebSocket state and writes its latest request/response back on drop

If a hook used that path, it could replace the continuation state for the real transcript. Starting fresh here and not caching on drop keeps hook inference from disturbing the next main turn.

[github-actions]

Closing this pull request because it has had no updates for more than 14 days. If you plan to continue working on it, feel free to reopen or open a new PR.

[abhinav-oai]

Moved to the internal stack: https://github.com/openai/codex-internal/pull/1279