Summary
External testing still reports manual pkill plus fresh ports are needed for ad-hoc local fixture servers. Leaked host servers block later runs and cause recurring friction.
Current wrapper state
- TOOL_CONTRACT documents that temporary HTTP servers launched outside the tool are host-owned.
- COMMAND_REFERENCE points extension maintainers to
startAgentBrowserContractFixtureServer() for deterministic localhost fixture servers instead of ad-hoc long-lived python3 -m http.server processes.
- SUPPORT_MATRIX tracks this as RQ-0115.
Ownership boundary
This native tool is intentionally a thin wrapper around upstream agent-browser. It does not allocate local ports, track arbitrary background PIDs, or clean up shell-started servers.
In-scope next action
Keep this as the local tracking link for future design work if fixture-server lifecycle becomes product scope. Any real wrapper-managed server feature needs a separate design pass with ownership, cleanup, security, and UX constraints; do not add an implicit server manager as part of routine browser navigation.
Summary
External testing still reports manual
pkillplus fresh ports are needed for ad-hoc local fixture servers. Leaked host servers block later runs and cause recurring friction.Current wrapper state
startAgentBrowserContractFixtureServer()for deterministic localhost fixture servers instead of ad-hoc long-livedpython3 -m http.serverprocesses.Ownership boundary
This native tool is intentionally a thin wrapper around upstream
agent-browser. It does not allocate local ports, track arbitrary background PIDs, or clean up shell-started servers.In-scope next action
Keep this as the local tracking link for future design work if fixture-server lifecycle becomes product scope. Any real wrapper-managed server feature needs a separate design pass with ownership, cleanup, security, and UX constraints; do not add an implicit server manager as part of routine browser navigation.