[realtonyyoung]

No GitHub or Linear issue

What & why

Start_agent_posts_once_and_returns_without_reading_the_agent_back failed whenever the suite ran from a coding agent, such as a Claude Code session in the Claude desktop app, while CI passed. The spawned kcap mcp flows finds KCAP_AGENT_ID and walks its ancestry for coding agents. It counts the desktop app and its claude as two, judges itself nested and sheds the id, so caller_agent_id went out null.

The test now starts its server through /bin/sh, which backgrounds the server and exits. The server is reparented to init, so its ancestry holds none of the suite's processes. The shell hands its pipes down, so the test talks to the server as before, and teardown kills the orphan by the pid the shell printed. On Windows the test spawns directly: Windows doesn't reparent, so the walk stops at the vanished parent anyway. No product code changes.

Verification

Run	Result
The test from a Claude desktop session, before	fails, caller_agent_id empty
Same, after	1/1 pass
McpFlowsServerTests	37/37 pass

🤖 Generated with Claude Code

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

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