Coding agents & CLI

• Claude Code v2.1.215: /verify and /code-review no longer auto-run on their own, invoke explicitly.
• Claude Code v2.1.214: fixes single-segment dir/** allow rules over-approving writes anywhere in the tree, closes a Windows PowerShell 5.1 permission-check bypass.
• Claude Code v2.1.212: /fork now spins the conversation into a background session with its own row in claude agents, old in-session subagent launch renamed /subtask.
• Claude Code v2.1.211: new --forward-subagent-text flag streams subagent text/thinking into stream-json output, permission previews now neutralize bidirectional-override/zero-width spoofing chars.
• stablyai/orca: ADE runs Codex/Claude Code/OpenCode/Pi side by side, each in its own git worktree, one dashboard with cross-provider usage tracking.
• esengine/DeepSeek-Reasonix: DeepSeek-native terminal agent as a single static Go binary, tuned around DeepSeek's prefix cache to keep long-session token costs low.
• alibaba/open-code-review: Alibaba's internal review bot open-sourced, deterministic pipeline + LLM agent, line-level comments, built-in NPE/thread-safety/XSS/SQLi ruleset, OpenAI/Anthropic-compatible.
• CLI that enforces agent governance with stop hooks + CI: locks agent behavior with stop-hook + CI gates instead of relying on AGENTS.md conventions alone.
• Termexo: local Windows workbench wrapping Claude Code and Codex.
• Lumichats: pitched as a Claude Code alternative for people who avoid the terminal.

LLM releases

• Kimi K3 2.8T-A50B: Moonshot's release, largest open model ever shipped, claimed Opus-4.8-class capability at Sonnet-5 pricing.
• GLM-5.2: flagged as a capability threshold, a real step-change for open agents.
• Thinky's Inkling: 975B-A41B multimodal, new best American Apache-2.0 open model, plus a 276B-A12B Inkling-Small variant.
• Latest open artifacts #21: Gemma 4, DeepSeek V4, Kimi K2.6, MiMo 2.5, GLM-5.1 all landed in one month, plus CAISI's independent V4 capability assessment.
• Latest open artifacts #22: Zyphra, Cohere, and Poolside broaden the open-model ecosystem beyond the usual frontier labs.
• Latest open artifacts #23: Laguna S2.1, Inkling, and Kimi K3 push open models further onto the capability/cost Pareto frontier.
• Claude Fable 5 and new AI safety fables: Anthropic swapped silent research-query manipulation for a classifier consistent with its other safety domains.

LLM mechanics & industry notes

• Some ideas for what comes next, May 2026: Gemini Flash 3.5, Mythos, the open/closed balance, and emerging power struggles as AI economics turn real.
• Frontier post-training recipe review with Finbarr Timbers: what an Olmo-style open recipe actually needs to reach frontier post-training quality.
• AINews: not much happened today: Codex is adding 1M users a day, Superapp usage keeps climbing.
• 5 Trends That Defined AI Engineering at World's Fair 2026: field shifted from building with agents to building systems around agents.

Vibe-coding & agentic web

• Panniantong/Agent-Reach: one CLI gives an agent read/search access to Twitter, Reddit, YouTube, GitHub, Bilibili, and XiaoHongShu without paid API keys.
• mvanhorn/last30days-skill: agent skill that researches a topic across Reddit, X, YouTube, HN, and Polymarket, then synthesizes a grounded summary.

Agentic SaaS

• SnapState: persistent state layer for AI agent workflows.
• diegosouzapw/OmniRoute: free MIT AI gateway, one endpoint to 290+ providers and 500+ models, quota-aware fallback, claims 15-95% token savings via prompt compression.