feather-performance-profiling

name	feather-performance-profiling
description	Use when an agent needs to profile frame time, memory, runtime overhead, hotspots, or performance regressions in a Feather session.

Feather Performance Profiling

Workflow

1. Establish a baseline:
   • Read feather://sessions/{id}/performance.
   • Note FPS, frame time, update/draw balance, memory, message volume, and Feather overhead.
2. Capture hotspots:
   • Use available profiler MCP tools or plugin actions before adding instrumentation.
   • Prefer short captures around the slow scene or interaction.
3. Correlate with logs and state:
   • Read logs and observer snapshots for scene transitions, entity counts, particle counts, asset loads, or shader changes.
4. Make the smallest performance edit:
   • Reduce allocations, repeated asset loads, excessive logs, expensive draw loops, or unbounded plugin payloads.
5. Re-measure:
   • Compare the same scenario before and after the change.

Guardrails

Do not optimize from one frame. Use repeatable captures, and call out when a result is noisy because the game, platform, or session changed.

References

• Read references/workflow.md for baseline metrics, capture strategy, common hotspots, fixes, and re-measurement format.