Skip to content

Kartik-Doye/Ravengard-AI-Recruiter / .agents/skills/hotpath

hotpath

Verified

Improves performance by defining a metric and workload, profiling first, controlling noise, changing one major variable, and comparing equivalent conditions. Use for latency, throughput, memory, CPU, rendering, query, bundle, or startup problems.

FileKindSize
SKILL.mdskill md11.5 KB
references/measurement.mdmarkdown3.1 KB
references/research.mdmarkdown2.4 KB
scripts/measure-reportother3.1 KB

references/measurement.md

View on GitHub
# Measurement toolbox: by layer, with the traps

Use what exists (project-native → installed tool → tiny bundled helper
→ recommend optional → manual). State the tool and conditions with every
number.

## Web frontend

- **Lighthouse / Core Web Vitals** (LCP, INP, CLS): lab numbers with
  throttling declared; field data (RUM) is truth when available.
- **DevTools Performance panel**: flame charts for long tasks, INP
  attribution, layout shifts.
- **Bundle analysis** (`webpack-bundle-analyzer`, `vite-bundle-visualizer`,
  `source-map-explorer`): what's actually in the payload; the 300KB
  "utility" nobody imports directly.
- **Network panel**: waterfalls: sequential requests that could
  parallelize, oversized payloads, missing compression/cache headers.
- Traps: lab throttle ≠ user devices; extensions skew runs; measure the
  production build, never dev mode.

## Backend / API

- **Query counting**: ORM logs, APM traces, or middleware timing; the
  N+1 shows up as 1 request → 200 queries.
- **Slow query log / `EXPLAIN ANALYZE`**: missing indexes and seq scans
  name themselves; compare planned vs actual rows.
- **APM/tracing** (OpenTelemetry, Sentry perf, provider APMs): p50/p95/
  p99 per endpoint; p99 is where the intermittently angry users live.
- **Load tools** (`autocannon`, `hey`, `k6`, `ab`): keep load below the
  saturation point you're trying to measure latency at, or numbers
  invert.
- Traps: local DBs with 12 rows lie about query plans; cold caches;
  connection-pool exhaustion masquerading as "slow DB".

## Application / language runtime

- **Profilers over timers**: cProfile/py-spy (Python), --cpu-prof/clinic
  (Node), pprof (Go), perf/flamegraph (Rust/C), JFR/async-profiler (JVM).
  CPU-bound or IO-bound stops being a debate.
- **Allocation/GC pressure**: most "memory leaks" are retention via
  forgotten listeners/maps/closures; heap snapshots diffed over time
  find the retainer.
- Traps: micro-benchmarks and JIT warmup; measuring debug builds;
  optimize-what-you-can-see (no profiler = guesswork).

## Data layer

- **Row counts and table growth**: the query fast at 10k rows dies at
  10M; test at realistic volume (seed it if needed: say so).
- **Index usage**: `EXPLAIN` every hot query; an index that exists but
  isn't used is a finding, not a plan.
- **Lock/transaction contention**: traces showing latency clusters at
  writes point here.

## CI / build / startup

- **Timing the phases**: `time` per step, build-tool profiling modes
  (`vite build --profile`, webpack stats), test-runner slow-report
  flags. Parallelize/cache the biggest phase, not the loudest one.

## Universal rules

1. Same conditions both sides of a change (machine, data, warmup, load).
2. Median/percentiles over single runs; report the spread.
3. The measurement method ships in the report: reproducible or it's an
   anecdote.
4. Realistic data volume: seeding up to production-like scale is part
   of the work, stated in the report.
5. **Regime Shifts over Micro-Tuning**: optimization breakthroughs come from structural changes (caching layers, data modeling, batching) not continuous granular parameter tweaking. Follow the trace, not the syntax.
slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack."Stars: 198kUpdated Claude CodeCodexVerifiedInstall
ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read.Stars: 194kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report.Stars: 194kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt.Stars: 194kUpdated Claude CodeCodexVerifiedInstall