zerg-testinglisted

# Zerg Testing ## Rules - Always use Make targets. Never run pytest/bun/playwright directly. - A proof run is incomplete while its provider process, Longhouse session, simulator/browser tab, scratch `HOME`, relay, or dev service is alive. - Create every test/QA session with an explicit hidden `launch_surface`; never use a human-looking project/title and rely on naming to hide it. Hidden launch provenance is not the same as user timeline hiding. - Before a QA cleanup receipt can pass, retire every exact Runtime Host session owned by the run: set timeline visibility to hidden, archive it, and verify the session ID is absent from the served agent inventory. Record `canary_session_hidden: true` in the receipt. `include_test=true` diagnostic surfaces can still expose a merely launch-hidden session. - Provider probes must use `--no-session` or an explicitly disposable `--session-dir`; never let a verification command write to David's default provider archive. If a protocol ignores that flag, record the exact native session path and remove it in the same `finally` block. - Every disposable proof run must allocate unique `mktemp -d` roots, record every path/PID it owns, and install an `EXIT/INT/TERM` trap; fixed `/tmp` names and untracked scratch files are prohibited. - Treat reparented descendants as still run-owned: a provider, bridge, shipper, catalog daemon, or verification server whose parent is PID 1 is not cleaned up. Record exact PID/PGID at launch,