Testing

Detail carved out of AGENTS.md, which keeps only what has to hold in every session. Repo-specific throughout; the facts here are verified against msb 0.7.6 and re-checked on 0.7.7.

• tests/ — stdlib unittest, run by check.sh: tests/harness.py (MsbctlTest) gives each test a throwaway HOME and a freshly loaded msbctl, with preflight() and the agent filter patched out and the network off; tests/fake/msb records argv and the env vars a test names, keeps state, and exits 2 on any call it does not know; tests/fake/tailscale records serve calls. Assert what reaches msb, and with assert_not_in_argv() that a planted secret does not.
• spikes/NN-*.sh — re-runnable PASS/FAIL scripts, one per mechanism the design relies on (SSH agent over vsock, mount ownership, per-port rules, persistence, hostname rules, memory reclaim, token placeholders, profile seeding, DNS rebind, a LAN model endpoint's key). Run one with ./spikes/07-claude-token-secret.sh. Several need host-specific env vars (e.g. ALLOW_HOST, ALLOW_PORT) with no defaults, on purpose. Each must assert both the allowed and the denied half.