Codex and Claude Code Changelogs

Codex fails to start because model_catalog_json in config.toml points to a missing file; deleting that line and model_provider = "magpie" is a temporary fix, with a root-cause fix and new release coming.

Claude Code 2.1.306 ships two CLI fixes: claude -p no longer reports 'Not logged in' in cloud sessions without env vars, and MCP-heavy sessions now show original API errors instead of 'prompt is too long'.

Claude Code 2.1.302 ships 63 CLI changes, including the Read tool refusing images over 128 MB, no prompt suggestions for credit-billed models after a credit-billed reply, and an idle compaction toggle in /config.

Claude Code released 2.1.299 and 2.1.300 back to back with 278 changes in two days, including auto-compaction, usage limit warnings in /usage, background task deduplication and sandbox escape fixes.

The Claude Code desktop update adds nested sessions that run tasks in parallel, proactive follow-up suggestions, a /btw tab for side questions, sidebar session status and recap cards, shifting the focus from writing code to managing multiple AI sessions.

OpenAI's Codex CLI is updated to 0.162.1, fixing a TUI crash on multi-line async questions, lost hyperlink destinations, and a startup issue from background server feature mismatches.

The new version adds an allow_large option to the Read tool so it can ingest a large text file in one call, and fixes redaction leaks in shared transcripts and debug logs.

OpenAI Devs says Codex Pro users can try composer predictions, which suggest the next message based on the conversation and how you talk to it.

The release makes failing or timing-out commands and HTTP hooks block execution, lets gateway upstreams limit the model list, and adds a stream startup timeout setting.

Codex CLI 0.162.0 adds managed Git worktrees for trusted local projects, task pinning in Command Center with p, and /copy for transcript blocks with clickable links.

Claude Code 2.1.294 fixes instruction-form prompt and agent hooks failing to block targeted commands, and improves Stop/SubagentStop hook judging to reduce premature agent stops.

OpenAI's Codex Cloud was quietly re-released, and the author says it is pretty good now.