Feature request: Claude-style /recap

Open 💬 6 comments Opened Apr 21, 2026 by yizhoupan3-debug
💡 Likely answer: A maintainer (github-actions[bot], contributor) responded on this thread — see the highlighted reply below.

What variant of Codex are you using?

CLI, CODEX APP

What feature would you like to see?

Two small but very high-leverage session UX features that Claude Code users already have:

  1. A built-in /recap command
  2. A /btw alias for the existing /side flow

I know Codex already has /side (thank you, that is useful). The main gap now is that Claude users often reach for two habits together:

  • /btw for a side question that should not pollute the main thread
  • /recap to quickly recover the state of a long-running session before continuing work

Why this would help

For longer sessions, resume/fork/return workflows are where Codex still feels heavier than Claude Code. The missing piece is not raw capability, but quick recovery of context and muscle-memory-friendly commands.

/recap would be especially helpful when:

  • resuming an older thread
  • returning after a long tool-heavy run
  • preparing to continue work after context compaction
  • checking current goals / constraints / next steps without asking for a full natural-language summary in the main thread

And /btw as an alias would help people coming from Claude Code immediately discover that Codex already supports the same side-conversation pattern via /side.

Proposed behavior

/recap

A manual command that produces a compact, structured summary of the current thread, for example:

  • current task
  • key constraints
  • important files / surfaces
  • decisions already made
  • next recommended step

Important: this should be a lightweight user-invoked recap, not only an automatic compaction artifact.

/btw

A simple alias for /side, with the same semantics:

  • open a side conversation
  • keep the main thread clean
  • return to the main thread without contaminating its working context

Related issues

  • openai/codex#18125 added /side for side conversations, which is great
  • openai/codex#16291 asks for better summary visibility around /resume
  • openai/codex#18637 asks for improved resume hint rendering
  • [#10823](<https://github.com/openai/codex/issues/10823>) shows long-session/context recovery is still a real pain point

Additional information

If maintainers prefer to keep only one command internally, /btw does not need separate implementation logic; an alias to /side would already solve the discoverability / migration problem for many Claude Code users.

The bigger request is really /recap: a fast, explicit “where are we and what should I do next?” command for the current session.

View original on GitHub ↗

6 Comments

github-actions[bot] contributor · 3 months ago

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

  • #18125

Powered by Codex Action

etraut-openai contributor · 3 months ago

Thanks for the feedback. I'm glad to hear someone is making use of the new /side command!

I'm reluctant to add an alias here. We've added aliases for other slash commands in the past, and it has resulted in confusion and negative feedback. I think it's preferable to have just one way to do something.

As for the /recap command, that sounds less like a command and more like a simple prompt ("Give me a concise summary of your progress" or something like that). If it's something that you use often, you could just save it as a skill that you can invoke using $recap or $progress. So I'm not convinced this needs a dedicated slash command.

xc1427 · 2 months ago

@etraut-openai Agreed on the /btw alias — one name per command keeps things clean.

On /recap though — I think you're right that the OP is asking for a command, and a skill like $recap could cover that use case. But it might be worth looking at this from a slightly different angle.

In Claude Code, recap is actually a semi-automatic background feature: when you've been idle 3+ min and your terminal is unfocused, it proactively shows a structured summary when you come back. No user action needed. The difference matters because a user-invoked recap — whether slash command or skill — burns tokens to regenerate state the system already had before compaction. An automatic one preserves and surfaces it at the right moment.

So beyond what the OP is requesting, there might be a deeper opportunity here: detect idle-return transitions and proactively surface a compact session summary. That's a session lifecycle feature, not a command surface thing.

For reference, Claude Code's implementation:

  • Triggers after 3+ min idle + unfocused terminal + at least 3 conversation turns
  • Never fires twice in a row
  • Skipped in non-interactive mode
  • On by default, toggled via config

The OP's /recap command could still exist as a manual fallback, but the automatic version is where the real UX win is.

joshka · 2 months ago
I'm reluctant to add an alias here. We've added aliases for other slash commands in the past, and it has resulted in confusion and negative feedback. I think it's preferable to have just one way to do something.

A possible approach to this is to name the commands internally (e.g. launch-ephemeral-sidechat or whatever) and then have config bindings that bind slash commands to internal commands. This approach is similar to what you'd do for custom keybinds that trigger commands.

One interpretation here of the user signal / observation here is that UX in a shell is often highly personal and configuratble (keyboard shortcuts, aliases, etc.). Codex looks a lot like a shell in that regard without the same affordances.

BobbyWang0120 · 21 days ago

@etraut-openai I'd like to reinforce @xc1427's point, because framing /recap as "just a simple prompt" undersells what makes it valuable. As a heavy Claude Code user, the value isn't the summary text itself — it's the trigger mechanism: it fires semi-automatically when you return to a session that's been idle and unfocused for a few minutes, surfacing a short structured recap of where things stand without you scrolling back through a long response or remembering to ask. A manual /recap command or $recap skill still puts the burden on the user to notice they've lost context and spend tokens regenerating it; the real win is the system detecting the idle-return transition and surfacing the state it already had, on its own. That's a session-lifecycle mechanism, not a command-surface thing — and it's the part I'd most love to see Codex catch up on.

t-kalinowski · 21 days ago

If this is implemented, it would be great if the recap was generated while the cache for the thread was still warm, to avoid a cold-cache hit just for the recap.

(related, it's be great if there was a setting we could configure to enable an automatic action like: "if cache is about to go cold and context is >70% used, compact").