BMAD Performance Optimization Skill

When to Invoke

• Reports latency, throughput, or resource regressions.
• Requests load/performance testing guidance or results interpretation.
• Needs to set or validate performance budgets and SLAs.
• Wants to plan scaling strategies ahead of a launch or marketing event.
• Asks how to tune code, queries, caching, or infrastructure for speed.

bmad-development-execution

Mission

Inputs Required

• Current architecture diagrams and deployment topology.
• Observability data: metrics dashboards, traces, profiling dumps, load test reports.
• Performance requirements (SLAs/SLOs, budgets, target response times).
• Workload assumptions and peak usage scenarios.

bmad-observability-readiness

Outputs

• Performance brief summarizing current state, key bottlenecks, and risks.
• Benchmark and load test plan aligning tools, scenarios, and success criteria.
• Optimization backlog ranked by impact vs. effort with owner and verification plan.
• Updated performance budget recommendations or SLO adjustments when necessary.

Process

1. Validate inputs and ensure instrumentation coverage. Escalate gaps to observability skill.
2. Analyze telemetry to pinpoint hotspots (CPU, memory, I/O, DB, network, frontend paint times).
3. Assess architecture decisions for scalability (caching, asynchronous workflows, data partitioning).
4. Define performance goals and acceptance thresholds with stakeholders.
5. Create load/benchmark plans covering baseline, stress, soak, and spike scenarios.
6. Recommend optimizations across code, database, infrastructure, and CDN layers.
7. Produce backlog with measurable acceptance criteria and regression safeguards.

Quality Gates

• Recommendations trace back to observed data or projected workloads.
• Each backlog item includes measurement approach (before/after metrics).
• Performance budgets and SLAs updated or reaffirmed.
• Risks communicated when goals require major architectural change.

Error Handling

• If telemetry contradicts assumptions, schedule hypothesis-driven experiments rather than guessing.
• Flag when performance targets are unrealistic within constraints; propose trade-offs.
• When required tooling is unavailable, document blockers and coordinate with observability & dev skills.