Fetching from the wire…
Tools2026-09-03 · source-backed
Released September 3, features.context_management.experimental_mode is off by default and, when enabled for eligible ChatGPT Plus/Pro/Pro Lite sessions on the Codex backend, activates token-budget context, history notes, and a new_context tool letting the model request a fresh window without spending tokens on a compaction summary. API-key sessions, custom providers and temporary structured threads are excluded. Every other harness bets on better summarization; this bets that a clean window beats a compressed one. Same release adds remote plugin marketplaces (#42150) and makes Guardian review history survive compaction, restarts and user-created forks.
Each link below shares sources, entities, or timing with this story.
The July 18 release notes bundle fixes that restore the full window, which means it had been silently degraded for some unspecified period. If you benchmarked those models in Codex over the past few weeks and found long-context performance underwhelming, you may have been meas...
The result-interception hook (#41202) is a place to redact secrets or strip injected instructions out of untrusted MCP output before it enters context, which is the missing seam in every MCP setup I've built. Also: a configurable grace period for discovering tools from optiona...
Three things happened this month that only make sense together. Agent Plugins 1.0 shipped co-signed by six competitors: AWS, Anysphere, Microsoft, OpenAI, Vercel and Google (GitHub Changelog). It makes skills-plus-MCP bundles portable across clients. OpenAI's August 11 Codex c...
upstash/context7 (60,590 stars) shipped @upstash/[redacted] on August 7 on the 2026-07-28 protocol revision. HTTP serving is now stateless for both modern and legacy clients, and Redis-backed sessions are gone, which is a real operational simplification for anyone self-hosting...
AMAP-ML released it August 4 (463 stars, v0.1.3 on August 7) on a strict Manager/Executor/Auditor split where, in its own words, "only results that pass independent verification enter persistent task state" (GitHub). WeaveBench 51.8% → 80.7% completion. OSWorld 2.0 2.8% → 8.3%...
Three repos in this week's data exist purely to run many coding agents at once: superset-sh/superset at 12,526 stars ("run an army of Claude Code, Codex etc."), agent-of-empires at 2,854 stars with both a TUI and a mobile-accessible web UI, and ruvnet/ruflo at 65,384 stars as...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.