Feature Request: Execution Replay / Debug Replay Mode

What variant of Codex are you using?

CLI

What feature would you like to see?

Feature Request: Execution Replay / Debug Replay Mode

Summary

Codex already provides:

• session rollout persistence
• event streaming
• sandboxed execution
• autonomous "exec" mode
• tool orchestration

However, there does not currently appear to be a structured way to replay or inspect a previous execution flow for debugging and reproducibility purposes.

A replay/debug mode could significantly improve investigation of:

• flaky autonomous runs
• sandbox-related failures
• unexpected tool behavior
• approval flow debugging
• session restoration issues
• CI/non-interactive execution diagnostics

---

Proposed UX

Examples:

codex exec --record "fix failing tests"

codex replay SESSION_ID

codex replay rollout.jsonl

Possible capabilities:

• replay event streams
• inspect tool calls in order
• inspect command outputs
• inspect approvals/denials
• compare generated diffs
• inspect failure points
• optionally re-run tools in dry-run mode

---

Why this could be valuable

Codex already has strong runtime foundations:

• session persistence
• event-driven architecture
• sandbox policies
• autonomous execution ("exec")
• MCP interoperability

Replay/debug tooling feels like a natural extension of the existing runtime model and could improve:

• reproducibility
• debugging workflows
• CI diagnostics
• trust/inspection of autonomous execution
• developer ergonomics for long-running tasks

---

Additional thoughts

Even a lightweight first version focused purely on event/timeline inspection (without deterministic model replay) would already be extremely useful.

For example:

codex replay --inspect SESSION_ID

to inspect:

• event order
• tool execution timeline
• command outputs
• approval decisions
• generated patches
• failure boundaries

Additional information

_No response_