fix(cloud): a new cloud chat opens even when its project's setup fails, and says so

[andrewcai8]

A new cloud chat failed to open when one of its repository's prepare commands failed ("Environment setup failed: Remote preparation failed: Preparation command failed ..."). Since #198 a chat that had prepared once tolerated a failing setup, but a chat's first preparation stayed fatal. So when megpt-mono's setup broke upstream (a query generated against a production Gel migration that has not shipped yet), no new chat could open at all.

Every preparation the manager makes for a chat now sends forChat to the guest. A chat's failing prepare commands are then handled the way a prepared root's already were. The remaining commands still run, and the masked failures join refreshError (logged on the host as a warning) and setup-failure.log in the box's T3 home. The T3 server starts, and the agent's cloud machine note quotes the log. forChat is left out of the guest's intent hash, like setup.

Warm base and spare builds stay fatal, so a broken setup is never sealed into something later chats share:

• preparing in EnvironmentControl leaves forChat off when the request is the build in progress that its repository's warm base (E2B) or spare (Namespace) record names (isBaseBuild in warmBases.ts). That record is saved before the build is driven, and only that build is ever sealed.
• The E2B seal rehearsal prepares from the raw stored manifest, so it never carries forChat.
• Namespace Mac template builders pass builder, which already drops the chat's setup. They now drop forChat too.

The guest's default is still fatal, so any preparation path that misses the flag fails the safe way.

Tests:

• remotePreparation.test.ts: a new chat whose first setup command fails opens with refreshError, writes setup-failure.log, runs the next command, and starts its server. Committed first and failing (Remote preparation failed: Preparation command failed: deps broke). The existing fatal test, renamed to cover warm base and spare builds, still refuses on every retry. The prepared-root tests from fix(cloud): a chat whose project setup fails still gets T3 updates #198 are unchanged.
• warmBases.test.ts: isBaseBuild against real stores tells the spare build from a chat, from the same request on the other provider, and from a request with no repository.

Claude Opus 5.5 via Claude Code.

🤖 Generated with Claude Code

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

ℹ️ The exact PR base did not have a successful artifact. Baseline uses the latest successful main measurement shown below.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	5.0 KiB	5.0 KiB	0 B (0.0%)	6.8 KiB	✅
Codex	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Codex	Live turn WebSocket wire	1.2 KiB	1.2 KiB	0 B (0.0%)	2.0 KiB	✅
Codex	Live turn WebSocket decoded	20.9 KiB	20.9 KiB	0 B (0.0%)	29.3 KiB	✅
Codex	Live turn messages	2	2	0 (0.0%)	8	✅
Claude	Total thread wire	5.0 KiB	5.0 KiB	0 B (0.0%)	6.8 KiB	✅
Claude	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Claude	Live turn WebSocket wire	1.2 KiB	1.2 KiB	0 B (0.0%)	2.0 KiB	✅
Claude	Live turn WebSocket decoded	21.2 KiB	21.2 KiB	0 B (0.0%)	29.3 KiB	✅
Claude	Live turn messages	2	2	0 (0.0%)	8	✅

Baseline: 57fde0d · PR result: e347d4e · Source CI: cancelled

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 108.5 KiB
• Claude decoded thread snapshot: 108.8 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.