KWDB

Use when CI or local verification fails in this KWDB repository and the first goal is to classify the failure, identify the most likely module, pick the first files to inspect, and choose the smallest reproduction command before attempting a fix

KWDB-specific code review skill. Reviews current git changes (C++ / Go) focusing on database-specific risks: persistence correctness, concurrency safety, SQL/TS semantics, cross-module contracts, error code mapping, compatibility, and performance regression. Triggers on: code review, PR review, code quality check, reviewing changes. Covers: Go (kwbase/) and C++ (common/ + kwdbts2/) code review.

Use when the user requests writing functional specifications, design documents, PRDs, technical design documents, or needs to analyze KWDB database requirements.

Use when the user asks to investigate KWDB performance via TSBS workloads. Routes to tsbs-benchmark for standard benchmarks, and coordinates tsbs-benchmark + perf-event-profiling pipelines for hotspot analysis. Currently only supports TSBS-based performance analysis. Do not use for general JDBC tests, SQL correctness tests, or non-TSBS performance issues.

Code refactoring for the KWDB database repository (Go/C++). Covers planning, complexity reduction, code smell elimination, and behavior-preserving verification.

MUST USE this skill for ANY Go unit test task in KWDB — even a simple "generate a _test.go" or "add tests" request. Covers file naming, 单元测试 conventions, package foo vs foo_test, leaktest goroutine detection, table-driven give/want tests with t.Run, testutils.IsError error assertions, testutils.SucceedsSoon async helpers, skip.UnderShort/UnderRace, TestMain + serverutils.InitTestServerFactory, StartServer for in-process SQL tests, testcluster.New for multi-node cluster tests, benchmark templates, testdata directories, and Makefile targets (test/testshort/testrace/stress/bench). Trigger whenever user says 单元测试/单测/测试文件/测试/unit test, references _test.go files, asks about testing conventions, or needs to write/generate/add tests for any Go code under kwbase/. Even for trivial one-file requests, the skill enforces leaktest, table-driven patterns, and proper error assertions. Covers kwdbts2 Google Test conventions as well.

Use when profiling a running server process with Linux perf — capturing performance events (cycles, cache-misses, instructions) while a client script triggers the workload, then analyzing perf.data for hot functions and hardware event metrics. Profiling and analysis can be invoked independently or combined.

Use when running KWDB local TSBS benchmark workflows. Covers two entry points: the full-pipeline qa/run_tsbs_test.sh for standard benchmarks, and the scripts under .agents/skills/tsbs-benchmark/scripts/ for workload setup (called by the kwdb-perf-investigation orchestrator). Do not use for general JDBC tests, SQL correctness tests, or non-TSBS performance issues.