Support Slash Commands in `codex exec` Mode

[jtele2]

What feature would you like to see?

Slash commands do not work in exec mode (or do not work in the way users expect).

For example, say I have a custom COMMIT slash command. In Claude Code, I'm able to run the equivalent of this easily: codex e "/COMMIT" - and it will run my slash command. Codex CLI does not seem to parse that as a slash command (and inject my prompt), but rather just interpret the raw text.

---

Proposal

Enable alias-slash commands (or a subset thereof) within the codex exec / codex e mode. That is, allow the user to prefix the prompt or combine slash commands with the prompt so that non-interactive runs can:

• Adjust or set the approval mode (e.g. --auto-edit, --full-auto), OR via slash command semantics
• Inspect status (e.g. /status)
• Possibly change model or reasoning level (via /model)
• Apply other interactive commands that should logically function even without full TUI

Examples

codex exec "/mode full-auto; fix the CI failure" (builtin)
codex exec "/approvals suggest; lint and format code" (builtin)
codex e "/status; run all tests" (builtin)
codex e "/COMMIT" (custom)

---

Motivation

• Make codex exec mode more powerful and flexible without needing to drop into interactive mode.
• Enable scripting or automation workflows that depend on triggering different contexts (mode, permissions, model) in a single command.
• Reduce friction for users who know what they want and don't need interactive prompts.

Are you interested in implementing this feature?

No response

Additional information

No response

[HomyeeKing]

Hi @ae-openai and team,

I'd like to share a concrete use case for this feature that would significantly improve automation workflows.

Use Case: Programmatic Triggering

Currently, when using codex exec in scripts or CI/CD pipelines, we have to choose between:

1. Interactive mode - manually set mode/model each time
2. CLI flags - verbose and hard to remember (--full-auto --model gpt-4)

Example: Git Hook Integration

#!/bin/bash
# pre-commit hook

# Current approach - verbose and error-prone
codex exec --full-auto --model gpt-4 "review staged changes"

# Proposed approach - clean and intuitive
codex exec "/mode full-auto /model gpt-4 review staged changes"

Example: CI/CD Pipeline

# .github/workflows/codex.yml
- name: Auto-fix linting issues
  run: codex exec "/mode suggest fix eslint warnings in src/"

Example: IDE Integration

IDEs could generate commands like:

codex exec "/reasoning high explain this function" << 'EOF'
def complex_algorithm(data):
    ...
EOF

Benefits

1. Single string - easier to pass around in scripts
2. Self-documenting - the command shows intent clearly
3. Composability - can chain with other shell features
4. Familiar - matches interactive mode experience

Would love to hear thoughts on prioritizing this for the exec mode. Happy to help with implementation details or testing.