21 commits across five threads — modal portaling, offline search-quality harness, model pool 0428/0504 merge, admin activity log, Nano Only mode gate, gpt-image-2 resolution passthrough, and a Grafana Loki Typst report that drove a hard call on what to retry and what to drop
Four small migrations between five commits — action cache that survives Redis TTL, prompt editor split, backend-streamed zip downloads, and a credit pill that stops disappearing
A deep read of Anthropic's April 23 postmortem on Claude Code quality degradation, the three independent root causes, and what it means for engineers building on shared LLM infrastructure
A side-by-side read of Microsoft's 12+ lesson curriculum and a 100+ template catalog — which fits which learner, and what beginner content systematically leaves out.
Microsoft's AI-oriented quantitative investment platform qlib re-emerges as the open-source backbone for LLM-driven alpha mining via RD-Agent
A snapshot of three open-weight MoE releases from the first week of May 2026, compared on parameters, active experts, and quantization
A deep dive into five Claude Code skill and agent collection repos that surfaced together on 2026-05-10 and what it says about skills becoming the primary primitive for agent engineering
A side-by-side reading of MemPalace and Hermes Agent surfaces two competing primitives for agent memory — explicit graph-indexed recall versus emergent tool-mediated scratchpads
How Anthropic killed agentic misalignment by training on the why behind aligned behavior instead of just demonstrating the right action — and what it means versus OpenAI's test-time compute bet
A look at claude-code-os, a Linux Mint based LiveCD that boots into Claude Code and Codex CLI in under a minute, and what it signals about the rise of Claude Code distros
