Claude Code subagent workflow suddenly gets slow and error-prone
A workflow writes a spec and plan first, then has a subagent execute each task while a separate QA agent reviews the work. This setup used to work well, catching many mistakes. Recently, though, the subagent keeps making simple mistakes; QA catches them, but then more mistakes follow, creating a repeating cycle.
Each task now takes far longer than before — a spec expected to take about 10 minutes now takes over two hours. Something called 'omc' appears to be involved in the process and may be adding delay, though this isn't confirmed as the cause. There have been no recent changes to the CLAUDE.md (which stays slim and clean), no changes to skills, and no environment adjustments.
Most of the code is , so there's no legacy code needing that could explain the slowdown. Usage costs are climbing sharply as a result.
Key points
- Workflow: write spec/plan, subagent executes each task, separate QA agent reviews results
- Subagent errors have become frequent, creating repeated QA-catch-and-redo cycles
- Task time ballooned from ~10 minutes to over 2 hours for a small spec
- A component called 'omc' is suspected but not confirmed as the cause of delay
- No changes reported to CLAUDE.md, skills, or environment; usage/costs are spiking