Source: docs/PERFORMANCE.md
Performance
This project tracks router performance at script level.
Baseline metric (2026-02-13)
route_workflow.pyCLI cold process average around 264 ms/call in local Windows environment.- This includes Python interpreter startup overhead.
Benchmark command
python skills/codex-workflows/scripts/benchmark_router.py --iterations 10000
The benchmark compares in-process functions:
- baseline:
route_workflow.route - optimized:
route_workflow_fast.route
Target
- Keep
route_workflow_fastaverage in-process latency below baseline. - Preserve routing correctness against baseline on representative queries.
Performance guidance
- Prefer in-process router use when possible.
- Avoid repeatedly spawning Python for each single routing request.
- Keep routing rules data-driven (
routing_data.py) to simplify optimization.