Problem

The gateway behavioral test suite (src/gateway/**/*.test.ts, run via vitest.gateway.config.ts) does not run in CI on pull requests:

• The unit test job uses vitest.unit.config.ts, which excludes src/gateway/** (vitest.unit.config.ts:18).
• scripts/test-parallel.mjs only includes the gateway suite when REMOTECLAW_TEST_INCLUDE_GATEWAY === "1" (scripts/test-parallel.mjs:105) — which CI does not set.

The only gateway test that runs in CI is src/gateway/server.preauth-hardening.test.ts, in the dedicated test-gateway-preauth lane (.github/workflows/ci.yml:170). The CI comment states the reason directly:

The full gateway suite is gated behind REMOTECLAW_TEST_INCLUDE_GATEWAY=1 because of orthogonal pre-existing failures unrelated to this slice

So the entire gateway behavioral suite is dark in CI because of ~2 pre-existing failing tests — an all-or-nothing exclusion rather than skipping just the failing tests.

Impact

Two recently-merged gateway fixes have their only behavioral tests in this dark suite, so neither is protected by CI:

Both fixes passed CI only via type-check (tsgo) and build — not behavior. A future edit to either file (including a large merge from upstream) could break the behavior and pass CI silently. The chat.ts error-surfacing fix in particular was previously lost in a large merge and went undetected for months, precisely because its test was not running in CI and passed vacuously against the reverted code.

Proposed fix (quarantine-and-re-enable)

1. Identify the ~2 pre-existing failing gateway tests (the "orthogonal pre-existing failures" the CI comment refers to).
2. Mark only those with explicit it.skip / describe.skip and a comment linking back to this issue, so the failures are visible and tracked rather than silently swept under the env-var gate.
3. Run the rest of the gateway suite in CI on every PR — either by including src/gateway/** in the test job's path, or by adding a dedicated gateway lane (mirroring test-gateway-preauth) and wiring it into the aggregate CI job's required-results check (.github/workflows/ci.yml:280).

This converts a binary all-or-nothing gate into per-test quarantine and restores behavioral CI coverage for the rest of the suite, including the two fixes above.

Acceptance criteria

• The gateway behavioral suite runs in CI on every PR (not gated behind REMOTECLAW_TEST_INCLUDE_GATEWAY=1).
• The ~2 known-failing tests are explicitly skipped with a tracking reference — not silently excluded via the env-var gate.
• src/gateway/server-methods/chat.error-propagation.test.ts and src/gateway/server.agent.gateway-server-agent-b.test.ts execute in CI and pass.
• The aggregate CI job fails if the gateway suite fails (wired into the required-results check).

🤖 Generated with Claude Code