[Feat] Block environment verification with manage_tasks await

[roomote-roomote]

Opened on behalf of Daniel Riccio. Follow up by mentioning @roomote-roomote, in the web UI, or in Slack.

Related issue

No GitHub issue. Internal reliability work from Slack-owned investigation into post-create environment verification (soft agent poll loops were not reliable).

Why this PR exists

• A maintainer explicitly invited this PR in the linked issue or discussion
• I am a maintainer / this is internal Roomote work

Slice A of platform-owned environment verification: ownership of verification must not depend on model-driven get_summary polling while the parent turn can go Idle/Ready mid-boot. Follow-up review fixed misclassification of stopping/failed runs and required message success before readiness claims.

What changed

• Added Roomote MCP manage_tasks action await: server-side blocks/polls a task summary until it settles (Completed / Failed / Canceled / Ready / Idle / NeedsInput) or times out (default 25m, max 45m).
• Result is a compact text payload with terminal label, ready yes/no, timed-out flag, waited duration, and error summary so callers keep one mid-turn tool call instead of sleep loops.
• Settlement prioritizes latest run failed/canceled over aggregate completed, and keeps stopped/shutting_down runs active (not ready).
• environment-setup now does list_environments → launch verification → await → required get_messages success assertion → bounded env-definition repair (≤2 extra full cycles) → report.

How it was tested

• vitest for await-task settlement/timeout/classification (+ fail-over-completed and stopped phase cases)
• vitest for environmentSetupSkill contract strings
• Worker/package typecheck clean; pre-push lint/types/knip green

Checklist

• The PR title follows the repo convention: [Fix], [Feat], [Improve], [Refactor], [Docs], or [Chore] followed by a user-facing description
• This PR is small and scoped to one change
• pnpm lint and pnpm check-types pass locally
• I added tests or included a clear manual validation note above
• I removed secrets, tokens, private keys, and customer data from code, logs, and screenshots
• If this change should appear in the changelog, I ran pnpm changeset

[roomote-roomote]

No new code issues found. See task

• apps/worker/src/mcp/roomote-mcp-server/await-task.ts treats running runs in stopped or shutting_down as settled, then the fallback classifies them as Idle and ready. Those phases precede final failure/cancellation, so keep polling or return an explicitly non-ready outcome.
• apps/worker/src/mcp/roomote-mcp-server/await-task.ts checks aggregate completed before a latest run's explicit failed or canceled status. A failed/canceled resume can leave the task aggregate-completed, causing await to mislabel that run as successful; prioritize the latest terminal run status.
• packages/cloud-agents/src/server/workflows/skills/standard/environment-setup/SKILL.md permits Completed, Ready, or Idle without an error to verify an environment without requiring the verifier's final message. A run completes whenever the agent turn ends, including when it reports a startup blocker, so require a get_messages success assertion before reporting readiness.

_{Reviewed 51bbbc4}

[roomote-roomote]

Fixed all 3 review findings in 3e5cfd3.

• Keep running+stopped/shutting_down active (keep polling; no ready settle)
• Prefer latest run failed/canceled over aggregate completed
• environment-setup now requires get_messages success assertion after await; no Terminal-only readiness

Commit: 3e5cfd3