Two tests in packages/cli/src/cli/agent-relay-mcp.startup.test.ts fail on a clean checkout because they assert on telemetry context that is read from local machine configuration:
registers owned tools, prompt text, fleet tools, and strips execution metadata from tools/listpasses telemetry context through Agent Relay MCP clients
The assertion expects a fixed telemetry payload, but the received object carries this machine's real values:
+ "agentRelayOrgSlug": "khaliq-gant-s-workspace-d3f705aa",
+ "agentRelayUserId": "0403d3ba-55ba-4ee6-aeee-13403d9aeac2",
"baseUrl": "https://relay.example.com/",
Reproduction
On an unmodified checkout of main (b0151fff2):
npx vitest run packages/cli/src/cli/agent-relay-mcp.startup.test.ts
→ Tests 2 failed | 24 passed (26)
Both failures are present before any change and reproduce identically after one, so they are environmental rather than a regression. I confirmed this while working on #1436 — baseline 24 passed / 2 failed, and 34 passed / same 2 failed with that change applied.
Why it is worth fixing
A test whose result depends on whose laptop it runs on cannot gate anything. It will pass in CI where the values happen to be absent or stubbed, fail locally for anyone with a real workspace configured, and train contributors to ignore a red suite in that file — which is where a real regression would eventually hide.
The fix is presumably to stub the telemetry context (org slug, user ID, distinct ID) rather than letting the resolver read ambient configuration during the test.
Filed from the delegation-identity work; not fixing it here to keep #1436 scoped.
Two tests in
packages/cli/src/cli/agent-relay-mcp.startup.test.tsfail on a clean checkout because they assert on telemetry context that is read from local machine configuration:registers owned tools, prompt text, fleet tools, and strips execution metadata from tools/listpasses telemetry context through Agent Relay MCP clientsThe assertion expects a fixed telemetry payload, but the received object carries this machine's real values:
Reproduction
On an unmodified checkout of
main(b0151fff2):Both failures are present before any change and reproduce identically after one, so they are environmental rather than a regression. I confirmed this while working on #1436 — baseline 24 passed / 2 failed, and 34 passed / same 2 failed with that change applied.
Why it is worth fixing
A test whose result depends on whose laptop it runs on cannot gate anything. It will pass in CI where the values happen to be absent or stubbed, fail locally for anyone with a real workspace configured, and train contributors to ignore a red suite in that file — which is where a real regression would eventually hide.
The fix is presumably to stub the telemetry context (org slug, user ID, distinct ID) rather than letting the resolver read ambient configuration during the test.
Filed from the delegation-identity work; not fixing it here to keep #1436 scoped.