$ the-wire · showcase
codex exec gains remote app servers, WSL clipboard goes lightweight
By RepoJournal · Filed · About OpenAI · Composed from the cited sources · methodology
Codex's day was spent cutting work it did not need to do: a new remote mode for noninteractive runs, and three quieter fixes that stop the tool from reading transcripts, manifests, and PowerShell it never had to touch.
Support remote app servers in `codex exec` openai/codex
`codex exec --remote <ADDR>` now runs a noninteractive turn against an existing app server's configuration and credentials, starting a new thread or resuming one by UUID, with human-readable or JSON output. Bearer-token auth comes via `--remote-auth-token-env`; conflicting endpoints and unsupported agent config overrides are rejected.
Avoid legacy transcript reads in the agents overview openai/codex
The agents overview no longer replays a legacy thread's entire rollout just to fill in one preview: saved prompt and reply previews are fetched only for threads with paginated history. Legacy prompt previews now update from user-message notifications and survive refreshes and reverts, and the latest observed reply preview is retained when a thread closes.
Defer plugin manifest loading until marketplace entries are selected openai/codex
Marketplace listings used to read every plugin manifest before policy, product visibility, and path or plugin ID deduplication discarded entries. Validation and manifest loading are now separate, with filtering applied first, so that discarded plugins no longer cost a read or emit parse warnings for plugins that never appear.
Enable slash-command completion for unavailable threads openai/codex
Slash-command completion now stays available while you edit an unavailable thread's draft, so navigation, Tab completion, and submission of eligible recovery and local commands still work, custom submit keys included. The PR gives the example of typing `/sub` and pressing Enter to select `/sub