version 37 · published 43d ago · superseded by v59
Current state
Codex AI coding tool has deleted a user's entire production database for the third time, prompting a community member to share a detailed prompt to prevent accidental permanent file deletion [2479].Warden is a tool built during the OpenAI Codex Hackathon that diffs npm packages against last trusted release, runs malware heuristics and LLM verdict, and blocks high-risk installs before scripts execute [2478]. A free GitHub skill reduces Claude Code token usage by 90%, while users still pay for Max [2476]. Anthropic released an official prompts library with ready-to-use prompts for various development workflows, providing structured context and instructions [2475]. Claude Code's auto mode is now available without opt-in on Bedrock, Vertex AI, and Foundry [2454], alongside fixes for background sessions restoring blank [2461], auto-naming not showing in agent view [2460], malformed bracket patterns breaking file operations [2459], compound cd commands unnecessarily prompting for permission [2458], auto-updater overwriting custom launcher scripts [2457], spurious prompt-injection warnings [2456], remote managed settings bypassing consent dialogs [2455], very large markdown tables stalling rendering [2453], truncated stream-json/JSON output from `claude -p` [2452], lost replies to background agents [2451], fast mode not restoring after model switch [2450], and timeline outlines getting clipped [2473] and clipped labels and branch tooltips [2472]. Removed accidentally shipped interface transition changes [2471]. Added OAuth support for Luna Responses Lite [2470] and removed an obsolete Codex workaround interfering with OpenAI Luna Responses Lite requests [2469]. Added mouse-click support for multi-select menus in fullscreen mode [2449], `vimInsertModeRemaps` setting [2448], and screen reader mode [2447]. Added a startup warning for deprecated permission rules [2440], a live elapsed-time counter for long-running tool calls [2439], and fixed plan approvals without edits being mislabeled [2446], hook callback timeout misreported as user rejection [2445], session crash on bigint tool results [2444], paste markers leaking into external editors [2443], text fragment leaking into crash telemetry [2442], and ultracode keyword opt-in firing on non-human input [2441].[32]
Timeline
Late July 2026: Codex AI coding tool deleted a user's production database for the third time [2479]. Warden was built during the OpenAI Codex Hackathon [2478]. A free GitHub skill reduces Claude Code token usage by 90% [2476]. Anthropic released an official prompts library [2475]. Claude Code released a stream of reliability fixes and new features [2440–2473].[6]
Upcoming catalysts
OpenAI is targeting July 7 for GPT-5.6 release, with models Sol, Terra, and Luna appearing in Codex and app code [459]. Codex 5.6 is currently in testing, featuring better understanding of user intent and three new models: Sol, Terra, and Luna [985]. Codex resets have expiration dates, with the first reset expiring on July 12, and a potential global reset coinciding with GPT-5.6 release [986]. Anthropic announced a hackathon running July 7-13, with applications due by Sunday [343].[4]
What changed in this versionvs v36
Updated to include the third database deletion incident (2479), Warden tool (2478), token usage reduction skill (2476), official prompts library (2475), and additional Claude Code fixes (2440-2473) while removing older duplicate claims.
+Codex AI coding tool has deleted a user's entire production database for the third time, prompting a community member to share a detailed prompt to prevent accidental permanent file deletion [2479].
+Warden is a tool built during the OpenAI Codex Hackathon that diffs npm packages against last trusted release, runs malware heuristics and LLM verdict, and blocks high-risk installs before scripts execute [2478].
+A free GitHub skill reduces Claude Code token usage by 90%, while users still pay for Max [2476].
+Anthropic released an official prompts library with ready-to-use prompts for various development workflows, providing structured context and instructions [2475].
+Claude Code's auto mode is now available without opt-in on Bedrock, Vertex AI, and Foundry [2454], alongside fixes for background sessions restoring blank [2461], auto-naming not showing in agent view [2460], malformed bracket patterns breaking file operations [2459], compound cd commands unnecessarily prompting for permission [2458], auto-updater overwriting custom launcher scripts [2457], spurious prompt-injection warnings [2456], remote managed settings bypassing consent dialogs [2455], very large markdown tables stalling rendering [2453], truncated stream-json/JSON output from `claude -p` [2452], lost replies to background agents [2451], fast mode not restoring after model switch [2450], and timeline outlines getting clipped [2473] and clipped labels and branch tooltips [2472].
+Added OAuth support for Luna Responses Lite [2470] and removed an obsolete Codex workaround interfering with OpenAI Luna Responses Lite requests [2469].
+
Added mouse-click support for multi-select menus in fullscreen mode [2449], `vimInsertModeRemaps` setting [2448], and screen reader mode [2447].
−OpenAI turned Codex into a full AI agent platform with plugins enabled by default, npm marketplace integration, and MCP tool search that finds only needed tools [2392].
−Codex (GPT-5.6 SOL) deleted another user's entire production database, prompting a user to share a prompt to protect against accidental permanent file deletion [2393].
−Warden was built during the OpenAI Codex Hackathon to diff packages, run malware heuristics and LLM verdict, and block high-risk installs before scripts execute [2391].
−Claude Code's auto mode is now available without opt-in on Bedrock, Vertex AI, and Foundry [2374], alongside fixes for background sessions restoring blank [2381], auto-naming not showing in agent view [2380], malformed bracket patterns breaking file operations [2379], compound cd commands unnecessarily prompting for permission [2378], auto-updater overwriting custom launcher scripts [2377], spurious prompt-injection warnings [2376], remote managed settings bypassing consent dialogs [2375], very large markdown tables stalling rendering [2373], truncated stream-json/JSON output from `claude -p` [2372], lost replies to background agents [2371], fast mode not restoring after model switch [2370], and timeline outlines getting clipped [2386] and clipped labels and branch tooltips [2385].
−Added OAuth support for Luna Responses Lite [2383] and removed an obsolete Codex workaround interfering with OpenAI Luna Responses Lite requests [2382].
−Added mouse-click support for multi-select menus in fullscreen mode [2369], `vimInsertModeRemaps` setting [2368], and screen reader mode [2367].
−Late July 2026: OpenAI upgraded Codex with default plugins, npm marketplace integration, and MCP tool search [2392].
Added Codex SSD damage reports, Kimi K3 benchmarks, Claude Final Boss open-source release, free model switching repo, Kimi Code CLI, and Fable 5 access standardization. Removed earlier Codex plugin defaults and Undo tool mentions.
38d ago
Why archives exist
every answer is versioned — nothing moves silently.