Feature Request: Add SlashCommand tool-style auto invocation to Codex CLI

What changed

• Request Codex CLI to expose SlashCommand style metadata so the assistant can auto run custom /commands when instructions mention them, matching Claude's SlashCommand tool behavior. (https://docs.claude.com/en/docs/claude-code/slash-commands)

Why

• Custom prompts only run when the human triggers them; the CLI still treats assistant initiated /command text as plain chat, including in codex exec, which blocks scripted workflows. (https://github.com/openai/codex/issues/3641)
• Users have already filed related feature requests for richer slash command support, so end to end parity would unlock those efforts. (https://github.com/openai/codex/issues/913)

How to try locally

1. Place a command such as write-unit-test.md into ~/.codex/prompts/, restart Codex, and confirm it appears in the / menu. (https://tech-bridge-log.com/blog/codex-cli-latest-gpt5)
2. In a session, tell Codex "please run /write-unit-test before touching tests" and observe that it echoes the text instead of invoking the command. (https://github.com/openai/codex/issues/3641)
3. Compare against Claude Code, which auto invokes custom commands through the SlashCommand tool when instructions mention /name. (https://docs.claude.com/en/docs/claude-code/slash-commands)

Flags removed or added

• None.

Safety limits

• SlashCommand auto invocations should still honor per command permissions, including disable-model-invocation and allowed-tools, so shell access stays gated. (https://docs.claude.com/en/docs/claude-code/slash-commands)

Fallback behavior

• If the SlashCommand budget is exceeded or permissions deny it, Codex CLI should fall back to today's manual /command handling. (https://docs.claude.com/en/docs/claude-code/slash-commands)

Related issues

• #913 Add ability to add custom commands / tasks. (https://github.com/openai/codex/issues/913)
• #2885 Add argument substitution for slash commands. (https://github.com/openai/codex/issues/2885)
• #3641 Support Slash Commands in codex exec mode. (https://github.com/openai/codex/issues/3641)