Graph-based persistent memory for Claude Code

★ 7.5 · ai-tooling

add-mnemon is a Claude Code skill that adds persistent graph-based memory to an agent via mnemon: the agent surfaces relevant past context before each response and stores new insights after every turn. The skill installs the mnemon binary into the container image through the Dockerfile and wires `mnemon setup --target claude-code` into `entrypoint.sh`, so Claude Code hooks are registered on every container start. Memory is written to `/home/node/.claude/mnemon/`, which maps to a host-mounted directory and survives container restarts. Two Vitest integration tests are included — one asserting the Dockerfile install layer, another confirming the entrypoint invocation — to catch accidental removals during upgrades. The skill applies only to agent groups using the default Claude provider; groups configured with a different provider never invoke the Claude CLI, so mnemon hooks do not fire for them.