fix(broker): verify worker process before spawn success

[khaliqgant]

Summary

• hold the broker spawn acknowledgement through a short process-stability window
• return the wrapper's exit status and worker-log path when it dies during startup
• remove the rejected worker from the registry so node agent list cannot show a stale entry
• add unit and real broker integration coverage for failed and successful startup confirmation

This is a stacked draft on #1425. That base PR supplies the requested Bug A workspace-resolution fix and its regression coverage: explicit flag/environment selection wins the shared workspace ladder, while conflicting persisted enrollment state is surfaced rather than silently re-homing the node. This commit supplies Bug B.

Reproduction evidence

Before this change, spawning a nonexistent provider printed:

Spawned repro-missing (definitely-not-a-cli, pty).

The immediate agent list was empty, while the worker log contained failed to spawn wrapped command.

After this change, the same real broker/CLI path exits non-zero with:

agent 'repro-missing-fixed' process exited during startup (exit status: 1); see worker log .../repro-missing-fixed.log

The immediate agent list remains empty. A cat control spawn still reports success, appears in the agent list, and releases normally.

Validation

• env -u AGENT_RELAY_MCP_COMMAND cargo test -p agent-relay-broker --quiet — 853 passed, 4 ignored; integration crates also passed
• cargo clippy -p agent-relay-broker --lib --bins -- -D warnings
• cargo fmt --all -- --check
• targeted broker integration — missing provider rejected and absent from list
• workspace-resolution regression files — 6 files, 102 tests passed
• npm run typecheck
• npm run lint — no errors (existing warnings only)
• npm run format:check
• npm test — 1,733 passed; the two existing relayfile stale-daemon tests failed locally, matching the known base-branch issue

Stack

• Base: fix(cli): make workspace activation reversible and broker binding explicit #1425 (fix/workspace-restore-rebind) — Bug A
• This draft: Bug B process confirmation
• Retarget to main after fix(cli): make workspace activation reversible and broker binding explicit #1425 merges

[coderabbitai]

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 3a7b734c-7db6-4b93-bfcb-7943c1ffff83

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[khaliqgant]

Closing as a duplicate of #1429.

Both PRs point at the same branch fix/node-up-enrollment-precedence-and-spawn-reliability at the same head 5d691f212, opened 38 minutes apart on 2026-08-04. Neither has any review — the only comment on each is CodeRabbit's identical "Review skipped — Draft detected", so nothing is lost by consolidating.

Keeping #1429 because its description states the user-facing symptom directly (the CLI reporting Spawned X success while the spawn actually failed) rather than only the mechanism.

Worth flagging on the way past: both were left as drafts, so no bot has ever reviewed this branch, and it has sat untouched for two days. That fix is more valuable than it looks — the exact failure it describes cost real diagnosis time overnight on 2026-08-05, when a spawn onto finn-mini reported success and the PTY died instantly because the launchd PATH omitted the directory holding claude. With this merged, that would have surfaced as an immediate error instead of a healthy-looking agent that was never alive.

Also note the branch name mentions enrollment precedence, but the diff is spawn-acknowledgement only — it does not address #1432 (node up silently ignoring a fleet enrollment when a project workspace pin lacks enrolledNodeId). That issue still has no fix in flight.