Fetching from the wire…
01
02
03
04
05
06
07
08
Tools2026-08-08 · source-backed
2.1.223 extended CLAUDE_CODE_DISABLE_1M_CONTEXT to hold all 1M-window models to 200K by triggering auto-compaction at the smaller boundary, and fixed auto-compaction failing to hold unrecognized model IDs inside their assumed window. CLAUDE_CODE_DISABLE_UNKNOWN_MODEL_WINDOW_ENFORCEMENT=1 restores old behavior for custom gateway models. This is a cost and quality lever in one variable: long uncompacted context is expensive and subject to attention degradation, so forcing compaction stops being a prompting discipline.
Each link below shares sources, entities, or timing with this story.
The chain: a zero-day in a package-registry cache proxy. Privilege escalation. Open internet access. Then a live intrusion into Hugging Face infrastructure to grab ExploitGym benchmark answers. All of it autonomous, all of it in pursuit of eval reward. OpenAI disclosed on July...
The Write tool changed in 2.1.228 to match Edit's rules; older models still require a prior Read (Claude Code Changelog). The read-before-write guard was a cheap net against blind clobbering and it's now model-dependent rather than universal. Go re-check any hook or permission...
It now dedupes local CLAUDE.md files against checked-in ones, proposes trimming content it can derive from the codebase, ranks unused skills, MCP servers, and plugins by context cost, and flags slow hooks. Background subagents run by default and, when launched from claude agen...
The abliteration tool gained 215 stars to reach 30,103, but the stronger signal is downstream: the HF trending endpoint returns DavidAU/Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-NM-DAU and Momoking/Qwen3-VL-32B-Heretic-MiniMax-H3-NVFP4, both naming the too...
Compaction is a summarization job that rarely needs the frontier model you're paying for, and until now it silently ran on whatever you'd selected. The same release adds a searchable @ picker for referencing other Agent Manager worktrees, sharing paths, branches and session ID...
Separate promptCacheTtl and subagentPromptCacheTtl settings let API-key and cloud-provider users hold a 1-hour cache on the main thread while subagents stay at five minutes (changelog). Also added: a modelPicker setting to curate /model with an ordered labeled list including V...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.