Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/orlando-japan/claude-code-setting/performance-profilingnpx skills add orlando-japan/claude-code-setting --skill performance-profilinggit clone --depth 1 https://github.com/orlando-japan/claude-code-settingWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/orlando-japan/claude-code-setting/performance-profiling)<a href="https://agentmods.dev/skills/orlando-japan/claude-code-setting/performance-profiling"><img src="https://agentmods.dev/badge/skills/orlando-japan/claude-code-setting/performance-profiling.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00032 | $0.00867 |
| Opus 5 | $0.00016 | $0.00434 |
| Sonnet 5 | $0.00006 | $0.00173 |
| Haiku 4.5 | $0.00003 | $0.00087 |
Grade A, and why
performance-profiling scanned grade A with 1 finding against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 3d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Makes network callslowCapability
Not a fault in itself. Listed so you know the mod talks to something, and to what.
- **Request latency:** `curl -w '%{time_total}\n' -o /dev/null -s URL` How it starts
The opening of the file, as written. The whole thing — 96 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Performance profiling
Optimization without measurement is just adding complexity. Measure first, always.
Step 0 — Define "slow"
What does the user experience? What's the actual number, and what's the target?
- "Page takes 4 seconds to load" → specific.
- "Feels slow" → useless. Get numbers.
Without a target, you won't know when to stop optimizing.
Step 1 — Measure the baseline
Use the simplest tool that gives real numbers:
- Request latency:
curl -w '%{time_total}\n' -o /dev/null -s URL - Frontend: browser DevTools Network + Performance panels
- Backend: language-native profiler (cProfile, pprof, async_hooks, perf)
- DB:
EXPLAIN ANALYZE, slow query log
Measure:
- Steady state, not first request (JIT, cache warm-up).
- Multiple samples (p50, p95, p99). Average alone hides tail latency.
- Realistic inputs, not empty fixtures.
Write the number down. You'll need it for the "did it help" check.
Step 2 — Find the biggest contributor
Profilers lie by default when used wrong. Sanity-check:
- Flame graphs for CPU — look for the widest bar at the bottom.
- Trace spans for distributed systems — the long spans are the real cost.
- Allocation profile for memory / GC pressure.
- Query log for DBs — most wall time is usually one or two queries.
80/20 rule. The slowest 20% of your code consumes 80% of the time. Find that 20% and ignore the rest.
Step 3 — Form a hypothesis
State out loud: "I think X is slow because Y." If you can't articulate why, you're guessing.
Common hypotheses:
- N+1 queries. Loop that hits the DB per iteration.
- Missing index. Full table scan.
- Sync I/O in async code. Blocking the event loop.
- Over-serialization. Converting a 10MB blob that gets thrown away.
- Bad cache policy. Cache thrash, miss on hot keys, TTL too short.
- Chatty protocols. Ten round trips where one would do.
Step 4 — Change one thing, measure again
- Change one thing. Multiple changes simultaneously = can't attribute.
- Re-measure with the exact same method. Different method, different numbers.
- Compute the delta. p99 before vs p99 after. Not "it feels faster."
- If no improvement, revert. Don't leave speculative optimizations in the code.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 3d ago First seen · 96 lines · 32 tokens per session scan A 6c4ea937e58c
performance-profiling is a skill published in the GitHub repository orlando-japan/claude-code-setting (2 stars, last pushed 3mo ago), licensed MIT. It adds 32 tokens to every session and 867 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 1 finding (makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
performance-engineer
Enforces measure-profile-optimize order, preventing intuition-driven optimization of non-bottlenecks.
performance-analysis
Comprehensive performance analysis, bottleneck detection, and optimization recommendations for Claude Flow swarms.
ncu-cuda-profiling
Automated NCU (Nsight Compute) profiling workflow with full metrics collection and persistent storage.
performance-optimization
Optimizes application performance. Use when performance requirements exist, when you suspect performance regressions, or when Core Web Vitals or load times need improvement. Use when profiling reveals bottlenecks that need fixing.
performance-optimizer
Systematic performance profiling and optimization for Python and web backends — measure first, fix second, verify the fix.
performance-expert
Expert-level performance optimization, profiling, benchmarking, and tuning.