[john-the-dev]

Problem

tests/gateway-stall-exit-inflight-send.test.py (added in #4564) has failed on main since it merged at 957e30f. That keeps tsc + tests (clean install) and python standalone tests (5/9) red for every gateway PR (for example #5883 and #5885). CI run 40239176069, job 121307141571:

• FAIL: the exit waited for the slow answer (3.3s)
• FAIL: the accepted message was archived as delivered
• FAIL: the relay accepted the message exactly once (posts=0)

Cause

#5515 landed on main between #4564's last CI run (head 1f79824, green) and its squash merge. Since #5515, the bridge adopts the registry's owner_dm_room only after /v1/room {"op": "members"} lists just the agent and its owner. This test's fake relay answered every /v1/room post with {"ok": true}, so the bridge never adopted the owner DM and held proactive-200.txt. The bridge log shows it: holding owner-directed proactive-200.txt: no owner DM reading from the gateway yet (never the pinned room). No post was sent, so all three delivery assertions failed. The bridge behaves correctly; the fixture was stale.

Fix

The fake relay now answers the members op with the agent and the owner, matching the fixture in src/remote-gateway-bridge.test.py. This is a test-only change of 4 lines.

Evidence

• On main with the change: passes 3 times out of 3 (OK — 0 failure(s), posts=1).
• On main without the change: fails (exit 1). This is the regression control.
• At the old fix(gateway): a wedged-but-alive bridge exits so the supervisor can restart it #4564 head (1f79824): passes, which confirms that fix(gateway): an owner-private message goes only to a room of the agent and its owner #5515's interaction is what broke it.

Regression test: tests/gateway-stall-exit-inflight-send.test.py
Before fix: FAIL — 3 failures on main 957e30f (posts=0)
After fix: PASS — 0 failures, posts=1

🤖 Generated with Claude Code

[qingyun-wu]

Approving at 8bc9b57 (agent-assisted, Qingyun's Sutando seat). The mock relay now answers the bridge's /v1/room op: members probe (src/remote-gateway-bridge.py) with exactly the agent and its owner, the shape the owner-DM adoption check requires. So the bridge adopts the DM and reaches the in-flight send again. Ran locally: the test fails at the parent (3 failures, the same signature as main's shard 5: "no successful poll", posts=0) and passes at this head (posts=1, 0 failures). Test-only, with no production change.

[liususan091219]

Approving at 8bc9b57.

The diff is test-only: 4 added lines in tests/gateway-stall-exit-inflight-send.test.py, no production change. The fake relay now answers the bridge's /v1/room op: members probe (src/remote-gateway-bridge.py:217) with just the agent and the owner, the same shape as the fixture in src/remote-gateway-bridge.test.py:141. That lets the bridge adopt the owner DM and reach the in-flight send the test exists to check.

Verified locally in a scratch worktree, hermetic env (temp HOME and workspace, localhost mock relay):

• main 957e30f: fails, 3 of the 6 assertions (exit wait 3.3s, not archived, posts=0). The bridge log shows holding owner-directed proactive-200.txt: no owner DM reading.
• Head 8bc9b57: passes 3 of 3 runs, OK — 0 failure(s).
• Head with the 4 lines removed: the same 3 failures come back (posts=0). The fixture is what fixes it.

CI at this head: all 9 python standalone shards pass, and license/cla is success (1 context). No unresolved review threads. #5883 and #5885 both fail on this same test right now (#5883 also has a separate pool-restart failure).

A note on "main green": the latest main run also failed shards 1/9 (watch-tasks-stream-shutdown-leaves-inflight-claim) and 6/9 (create-worker-roster-preflight). Those files aren't touched here and pass at this head, so they look like flakes, but this PR doesn't fix them.

Second approval at this head: @qingyun-wu.

— Sutando (susans-MacBook-Pro)

[github-actions]

Coverage Gate

✅ Diff coverage PASSES the 95% bar. Whole-tree (informational): 94%.

Diff Coverage

Diff: origin/main...HEAD, staged and unstaged changes

No lines with coverage information in this diff.