performance-proof
/performance-proof
Find and validate a measurable performance bottleneck before optimizing. Use when latency, throughput, memory use, render speed, or compute cost is a concern.
- Author
- voyager67 ↗
- Repository
- AgentFieldkit ↗
- Licence
- MIT
- Last repository change
- 30.10.2026
Assessment and security
- Files in this package
- 1
- Scripts
- 0
- HTML comments
- 0
- URLs in SKILL.md
- 0
No signals detected. Repository changed on 30.10.2026, without references to retired models or APIs.
MIT, open: use and modify while retaining author attribution.
What it does and when to use it
Find and validate a measurable performance bottleneck before optimizing. Use when latency, throughput, memory use, render speed, or compute cost is a concern.
Original author description
Find and validate a measurable performance bottleneck before optimizing. Use when latency, throughput, memory use, render speed, or compute cost is a concern.
How it works
- Procedure
- Result format
- Boundaries
SKILL.md preview
--- name: performance-proof description: Find and validate a measurable performance bottleneck before optimizing. Use when latency, throughput, memory use, render speed, or compute cost is a concern. license: MIT --- # Performance Proof ## Procedure 1. Define the metric and target: p50/p95 latency, throughput, memory, startup time, bundle size, GPU utilization, or cost per job. 2. Record a baseline under representative inputs and a repeatable environment. Note warm-up, concurrency, dataset size, hardware, and variance. 3. Profile the workload using available instrumentation before changing code. Identify the bottleneck and its evidence. 4. Form one optimization hypothesis with a predicted measurable effect. Avoid speculative rewrites. 5. Make a localized change, preserving behavior and interfaces. If several changes are necessary, isolate them so regressions can be attributed. 6. Run the same benchmark and correctness tests. Compare distributions where possible, not a single favorable measurement. 7. Check trade-offs such as memory, complexity, tail latency, warm-up, quality, and operational cost. ## Result format Metric, baseline, environment, change, new measurement, variance or caveat, correctness check, and recommendation. ## Boundaries - Never call an optimization successful without comparable measurements. - Do not use production-scale load, incur cloud/GPU spend, or affect live customers without authorization. - If no reliable baseline can be obtained, present the change as an unverified hypothesis.
Versions
The database has no other skills named performance-proof.
Database snapshot: 13.11.2026 · package version 725ed56cd311 (SHA-256 digest of the entire package)
Downloads in the last 30 days: 0 · ranking
Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.
Similar skills
- 95performance-review
[Debugging] Use when a workflow step or the user asks for performance bottleneck analysis and validated fixes: queries, memory, network, rendering and concurrency. - 95performance-review
[Debugging] Use when a workflow step or the user asks for performance bottleneck analysis and validated fixes: queries, memory, network, rendering and concurrency. - 100performance-optimization
Use when performance is a real concern — measure first, isolate the bottleneck, and prove the improvement instead of guessing - 100performance-audit
Identify performance bottlenecks — rendering, startup, memory, and I/O — across any stack. Use when the app is slow, before perf-sensitive releases, or when the user mentions latency, jank, or memory growth. - 100review-performance
Review code changes for performance, resource use, scale assumptions, benchmarks, and latency or throughput regressions. Use for /review-performance. - 90make-it-fast-when-it-matters
Use when performance matters, something is slow, or a job is failing/timing out — measure before optimizing, profile to find the real bottleneck, parallelize independent work and set timeouts on anything that can hang; resist optimizing on a guess, but don't ship lazily slow either.