- Tests 7 failure modes: connection refused, server crash, timeout, 500 errors, invalid token, malformed JSON, consistency
- All agents gracefully degrade to Beads-only mode
- Uses --no-daemon flag for fast execution (33s vs 162s)
- Mock HTTP server simulates all failure scenarios
- JSONL consistency verified across all failures
Amp-Thread-ID: https://ampcode.com/threads/T-a1a3b39a-9002-44c8-8048-3292d894f404
Co-authored-by: Amp <amp@ampcode.com>