Agent Skills: Performance Optimization Workflow

Performance Optimization Workflow

1. Capture and Analyze

Make a request with tracing enabled — returns structured JSON with query counts, duplicates, issues, and span tree:

uv run plain observer request /path
uv run plain observer request /path --user 1
uv run plain observer request /path --method POST --data '{"key": "value"}'

The --user flag accepts a user ID or email.

The JSON output includes:

• response.status — HTTP status code
• trace.query_count / trace.duplicate_query_count — query summary
• issues — pre-detected problems (duplicate queries, exceptions)
• queries — each unique query with count, total duration, and source locations
• spans — nested span tree showing the request flow

2. Identify Bottlenecks

Check the issues array first — duplicate queries are flagged automatically with source locations. Then review:

• N+1 queries (duplicate queries with count > 1)
• Slow database queries (high total_duration_ms)
• Missing indexes
• Unnecessary work in hot paths

3. Apply Fixes

• Add select_related() / prefetch_related() for N+1
• Add database indexes for slow queries
• Cache expensive computations

4. Verify Improvement

Re-run uv run plain observer request /path and compare query_count, duplicate_query_count, and trace.duration_ms.