Featured

• 19 Aug

  Watching 5.5 million tokens move through one Claude Code session

  Eight questions, 56 API calls, 5.5 million tokens — tracing exactly where a Claude Code session's context budget goes, and why /compact only helps sometimes.

  9 min read
• 5 Aug

  Your agent's guard hooks don't cover its MCP tools

  A security gap in guarded AI coding agents: command-level hooks don't see MCP tool calls, even when the tool performs the exact same action.

  3 min read
• 4 Aug

  Starting the log

  This blog is live. Here's what it's for, and where the rest of the story is.

  1 min read