$ the-wire · showcase
Device-code sign-in for headless Codex, and audio that survives tool output
By RepoJournal · Filed · About OpenAI · Composed from the cited sources · methodology
Codex unblocked gateway authentication over SSH and stopped flattening file-backed audio into a text file ID, while its app-server settings APIs gained experimental environment selection and Cloud conversations got model and reasoning-effort pickers.
Support gateway device-code sign-in in headless TUI sessions openai/codex
Gateway sign-in now prefers device authentication in SSH sessions, headless Linux environments, or when no browser handler is available, and it completes the required gateway sign-in before requesting models, including on thread resume. The verification prompt is displayed in the terminal, so the browser round trip that blocked remote sessions is no longer required.
Preserve file-backed audio in tool outputs (#60735) openai/codex
Converting message content to tool output used to reduce file-backed audio to a text file ID, dropping its audio type; input_audio tool output items now accept either audio_url or file_id and preserve both. History estimates and output truncation account for file-backed tool audio and only inline audio data is prepared, while uploaded audio in raw function and custom tool outputs is rejected.
Limit MCP product attribution to configured REPL-backed calls (#60697) openai/codex
MCP request metadata could forward a client-supplied product_sku to unrelated servers and connectors; product_sku is now stripped from turn metadata before forwarding. Only Node REPL-backed servers and the browser connector repopulate it from the turn's apps_mcp_product_sku configuration, with unrecognized values mapped to other and empty or missing values omitted.
Support environment updates in app-server settings APIs openai/codex
turn/settings/update and thread/settings/update take experimental environments fields: turn updates apply to later steps of the running turn, thread updates to future turns, omission or null preserves the current selection, and [] clears it. The ordered environment selection is included in thread/settings/update.
Enable model and reasoning effort selection for Cloud conversations openai/codex
Cloud conversations used to take whatever model the activation returned with no discovery or change path from the TUI; the account-scoped catalog is now fetched on demand through /v2/models, independently of the submission and attachment queue, and a supported model and reasoning effort can be staged for the next new turn.