Performance Tuning (Deep Workflow)

Performance work is measurement-driven. Profile before optimizing; verify after changes; guard against regressions with benchmarks or production metrics.

When to Offer This Workflow

Trigger conditions:

• High CPU, memory, p99 latency, GC pauses
• Cost reduction via efficiency
• Premature optimization requests—need evidence first

Initial offer:

Use six stages: (1) frame goals & SLOs, (2) measure baseline, (3) profile & hypothesize, (4) implement changes, (5) verify & compare, (6) prevent regression). Confirm language/runtime and environment (prod-like data volume).

Stage 1: Frame Goals & SLOs