PrizmKit Performance Profiler

Identify performance bottlenecks through static code analysis and profiling guidance. Suggests targeted optimizations with expected impact.

Commands

prizmkit.perf-profile [module-or-file]

Analyze code for performance bottlenecks and generate optimization recommendations.

STEPS:

1. Read .prizm-docs/ for architecture and module relationships (dependencies, data flow, hot paths)
2. If specific file/module given: focus analysis on that scope; otherwise analyze project-wide critical paths
3. Analyze for common bottleneck patterns:
   • N+1 queries: ORM/database calls inside loops
   • Missing indexes: queries filtering or sorting on unindexed columns
   • Synchronous blocking in async code: blocking I/O in async handlers, missing await
   • Unnecessary serialization/deserialization: repeated JSON parse/stringify, redundant marshaling
   • Memory leaks: growing collections without bounds, unclosed resources, event listener accumulation
   • Inefficient algorithms: O(n^2) where O(n) or O(n log n) is possible, nested loops over large datasets
   • Missing caching opportunities: repeated expensive computations or external calls with stable inputs
   • Excessive logging in hot paths: string formatting and I/O in tight loops
   • Large payload transfers: over-fetching from database, sending unnecessary fields to client
   • Connection management: not reusing connections, missing connection pooling
4. Generate profiling recommendations:
   • Suggest profiling tools appropriate for the project's tech stack:
     • Node.js: --prof, clinic.js, 0x
     • Python: cProfile, py-spy, line_profiler
     • Java: JFR, async-profiler, VisualVM
     • Go: pprof, trace
     • Rust: perf, flamegraph, criterion
   • Provide specific commands to run the profiler against the identified hot paths
   • Identify measurement baselines to establish before optimizing
5. Output performance report:
   • Suspected bottlenecks: ranked by likely impact (HIGH / MEDIUM / LOW)
   • Suggested optimizations: for each bottleneck, concrete code changes or architectural adjustments
   • Profiling commands: copy-paste ready commands to validate each suspicion
   • Expected impact: qualitative assessment of improvement (e.g., "Eliminates N+1, expect ~10x query reduction for list endpoints")
6. Suggest updating .prizm-docs/ TRAPS section with discovered performance pitfalls for future reference

Path References

All internal asset paths MUST use ${SKILL_DIR} placeholder for cross-IDE compatibility.

Output

• Performance analysis report printed to console
• Profiling command suggestions ready to execute