Articles from gist.github.com
7 kept
- What xAI Grok Build CLI actually sends to xAI - a wire-level analysis (grok 0.2.93)
xAI Grok Build CLI sends file contents—including .env secrets—to xAI via model turns and uploads entire repositories as git bundles to GCS bucket grok-session-traces, regardless of what the agent reads.
- LLM Wiki v2 — extending Karpathy's LLM Wiki pattern with lessons from building agentmemory
Extends Karpathy's LLM Wiki pattern with memory lifecycle, knowledge graphs, hybrid search, and automated quality controls to keep personal wikis healthy as they scale.
- Claude for Chrome Extension Internals (v1.0.56)
The Claude for Chrome extension uses a Manifest V3 architecture with a React side panel, an agentic tool-use loop, and extensive Chrome DevTools Protocol integration for AI-driven browser automation.
- Research After AI: Principles for Accelerated Exploration
The document outlines principles for responsible AI use in research, emphasizing judgment, deliberate pacing, and shared standards over speed or output volume.
- Claude Code Multi-Agent Orchestration System
Claude Code v2.1.19 contains a hidden multi-agent orchestration system called TeammateTool, confirmed via binary analysis, enabling coordinated swarms of specialized AI agents.
- CLAUDE_AGENT_SDK_SPEC.md
The Claude Agent SDK specification defines a JSON-message protocol for programmatic access to Claude Code's agentic capabilities via stdin/stdout.
- rl-wrong-about-rewards.md
The standard reinforcement learning formalization errs by placing reward in the environment; instead, reward should be part of the agent, enabling goal-driven behavior.