host tests: wait for new-room skills to load before simulating bot message

[habdelra]

Summary

Two acceptance tests in commands-test.gts intermittently failed on CI (shard 17 of the host test suite):

• ShowCard command added from a skill, can be automatically executed when agentId matches — timed out waiting for apply-state="applied"
• command request with arguments that do not match the json schema gets an "invalid" result — warning text was No command for the name "show-card_566f" was found instead of the expected JSON-schema validation message

Root cause

Both tests click [data-test-create-room-btn] and then immediately simulateRemoteMessage with a command request for show-card_566f, which is a function on the env skill that the new room auto-attaches.

The new room's processRoomTask is still mid-loadSkills when the simulated bot event lands; message-builder walks builderContext.skills and calls store.get<Skill>(skill.cardId), but the env skill is not in the store yet, so the request never resolves to a codeRef and command-service falls into the "No command for the name X was found" branch instead of executing or running the JSON-schema validator.

Fix

• New helper waitForNewRoomSkillsLoaded(roomId): polls until the room's enabledSkillCards is populated and roomResource.commands.length > 0 (commands is computed from the loaded skill instances in the store, so a non-empty list proves loadSkills finished and the skill's @field commands field is materialised). On timeout it throws with a JSON diagnostic snapshot of skillsConfig + the room resource's view, so a future flake reports enough state to distinguish "no skill loaded" from "skill loaded but commands missing".
• Used in the two failing tests and the third sister test (is not automatically executed when agentId does not match), which has the same setup and was passing by luck.
• Inline diagnostic logging in the two failing tests around the assertions that have been flaking, so we capture rendered apply states, warnings, and room/skill state if either still misses after the fix.

Test plan

• CI passes shard 17 of host tests
• Run the three affected tests locally a few times to confirm no flake
• If a CI failure recurs, the diagnostic logs in the test/helper attribute it cleanly

🤖 Generated with Claude Code

[github-actions]

Preview deployments

• Host staging preview

• Host production preview

Host Test Results

    1 files      1 suites   1h 49m 0s ⏱️
2 724 tests 2 709 ✅ 15 💤 0 ❌
2 743 runs  2 728 ✅ 15 💤 0 ❌

Results for commit 00a61ed.

Realm Server Test Results

    1 files      1 suites   10m 5s ⏱️
1 482 tests 1 482 ✅ 0 💤 0 ❌
1 573 runs  1 573 ✅ 0 💤 0 ❌

Results for commit 00a61ed.

[Copilot]

Pull request overview

Reduces CI flakiness in host acceptance tests by ensuring a newly created AI assistant room has finished loading its default skills (and derived commands) before simulateRemoteMessage dispatches command requests that depend on those skills being present in the store.

Changes:

• Added waitForNewRoomSkillsLoaded(roomId) test helper that waits until the room’s enabled skills are present and the room resource exposes non-empty commands, with enriched timeout diagnostics.
• Updated three acceptance tests to await skill/command readiness before simulating show-card_566f command requests.
• Added targeted console diagnostics on the previously-flaking assertions to improve future failure triage.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
packages/host/tests/helpers/index.gts	Adds a polling helper that waits for new-room skill loading to complete and enriches failures with diagnostic state.
packages/host/tests/acceptance/commands-test.gts	Uses the new helper to remove the skill-load race before simulating bot command requests; adds diagnostic logging around flaky assertions.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.