Session Extractions
How multiple LLMs, skills, and agents share a project without losing the thread — a shared markdown brain, a conflict-free handoff protocol, and a distributable Claude Code plugin.
Teaching Blocks
The architecture behind a context manager that lets Claude, Codex, Gemini, and other CLI windows hand off work across sessions — and why the right frame is agent-to-store, not agent-to-agent.
Read →How a working multi-agent context manager got built across one session without a single written requirement — driven entirely by five interrogation moves: grill the forks, stress-test as a thought experiment, walk the workflow frame-by-frame, question the package, count your copies.
Read →The same coordination protocol enforced three ways across Claude, Codex, and open-source models — and why pretending they're equal is the real bug. Harness hooks vs. instruction files vs. system prompts, and how a shell wrapper narrows the gap without erasing it.
Read →How to ship AI capabilities so they're available everywhere you want them and invisible everywhere you don't — project-local plugin scope, bundling what always travels together, and why a slash command beats a skill for one-shot install operations.
Read →Skills
The Claude Code plugin that installs the shared brain into any project — .context/, hooks, routing table, CLI bootstraps. Bundles the context-mgr installer skill and the session check-in/check-out runtime. Download includes the full plugin folder.