What problem does it solve? Benchmarking individual KHI inspection tasks is difficult because they run inside a concurrent DAG with external dependencies like the Cloud Logging API, making measurements noisy and slow. This Skill provides recipes to isolate a single task, record its upstream inputs as fixtures, and replay them for fast, repeatable microbenchmarks and pprof profiling. ## Core Features & Use Cases - Record & Replay Benchmarks: Capture live upstream task output as JSON fixtures with KHI_RECORD_TASK_RESULTS=1, then replay them offline via JobTestHarness without GCP credentials. - Task Isolation: Configure JobTestConfig with RecordedTasks and TargetTask to benchmark exactly one task (log ingester, log grouper, or timeline mapper) without downstream execution. - CPU & Memory Profiling: Generate pprof CPU and heap profiles scoped strictly to the target task using KHI_TASK_CPUPROFILE and KHI_TASK_MEMPROFILE. - Use Case: You want to reduce allocations in the Kubernetes audit log ingester. Record the upstream log query task once, then run go test -bench locally to measure ns/op and allocs/op improvements after each optimization. ## Quick Start Ask the AI to create a benchmark for a specific KHI task using the JobTestHarness boilerplate, then record fixtures and run the benchmark with profiling enabled.
