Add Mnemon Persistent Memory

What it does

Installs the mnemon binary into the NanoClaw agent container image and wires mnemon setup into the entrypoint. On every container start it registers Claude Code hooks that surface relevant memory before the agent replies and store new insights after each turn.

The runbook covers:

• Adding the mnemon install layer plus MNEMON_DATA_DIR to container/Dockerfile
• Inserting mnemon setup --target claude-code --yes --global into container/entrypoint.sh, with output routed to stderr so it never breaks the JSON stdin handshake
• Copying two structural tests that guard the Dockerfile layer and entrypoint wiring, then running them with vitest
• Rebuilding the image, restarting the service, verifying hooks, and locating the host memory path

Memory lands in the per-agent-group .claude/ mount, so it survives container restarts.

Who it's for

• NanoClaw operators tired of re-explaining context in every new agent session
• Agent groups on the default Claude provider (no provider key, or "provider": "claude")
• Users comfortable editing Dockerfiles, reading docker logs, and restarting systemd/launchd units

Examples

1. Pre-flight provider check: run grep -H '"provider"' groups/*/container.json to exclude groups on other providers (e.g. opencode), where mnemon hooks never fire.
2. Pin the latest release: fetch the newest tag from the GitHub releases API, set it as MNEMON_VERSION, rebuild with ./container/build.sh, then smoke-test docker run --rm --entrypoint mnemon nanoclaw-agent:latest --version.
3. Verify recall: chat with the agent, open a fresh session, and reference something from before. If mnemon surfaces the context without you restating it, the hooks work; to reset, stop the container and delete the mnemon/ subdirectory on the host mount.

Install the skill I found on Claude Skill Mart.
Copy the .claude/skills/add-mnemon folder from the GitHub repo nanocoai/nanoclaw into my ~/.claude/skills/add-mnemon/.
When it's done, tell me in one line what this skill can do.