mirror of
https://github.com/EveryInc/compound-engineering-plugin.git
synced 2026-09-19 01:09:55 +08:00
ff63cb83c0
peer-job-runner.test.ts drives real detached processes and real bounded waits, and ran on bun's 5000ms default per-test timeout. Its bounded-wait case measured 4653ms on CI -- a 347ms margin -- so whether it passed was becoming a question of how busy the runner was, not whether the runner script is correct. Parallel workers narrow that margin further. Match the 20s headroom the other subprocess-heavy suites already take. Also record why the worker count stays unpinned: --parallel=8 was measured on CI and buys ~9% wall time while inflating total test-CPU 223s -> 343s and pushing five tests past the 5000ms default. The remaining wall time is bounded by the slowest single file, not by worker count. Claude-Session: https://claude.ai/code/session_01EW4bpYuKU2KpVy3qqEZX9r