[Bug] supervisor-owners registry ignores PRIME_AGENT_CODING_AGENT_DIR — isolated test runs leave stale owner records in the real ~/.prime (0.8.1) #1889
KoLyslaV
started this conversation in
Bug reports
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Affected area
Daemon (supervisor ownership registry) / test isolation
Environment
packages/coding-agent) on a machine that also runs a real daemonWhat happened?
A test-run supervisor started with a fully isolated agent dir still writes its ownership record into the real user-level registry
~/.prime/supervisor-owners/.Repro observed with:
Afterwards the real registry contains e.g.
~/.prime/supervisor-owners/<generation>.owner/owner.jsonwith:{ "socketPath": "/var/tmp/<isolated>/tmp/prime-supervisor-<pid>-<hash>.sock", "agentDir": "/var/tmp/<isolated>/tmp/prime-daemon-supervisor-test-<rand>/agent", "phase": "owner" }i.e. the record honors the isolated
agentDirin its content but not in its location. If the test run is killed (CI abort, SIGSTOP/SIGKILL), the stale owner record stays in the production registry forever.A second, worse variant: running the same suites without
PRIME_AGENT_CODING_AGENT_DIR(easy to do from docs) had test supervisors log into the real~/.prime/agent/logs/*and touch the real/tmp/prime-agent-<uid>/update-restart-coordinatorswhile a production daemon was live on that socket dir.Impact
Suggested fixes
supervisor-ownersroot from the same agent-dir resolution as everything else (soPRIME_AGENT_CODING_AGENT_DIRisolates it too), or honor an explicit override env.All reactions