perf-reviewlisted

# Performance review Gates a **code change against its performance bar** by running it — five measured signals: **latency & throughput** (tail incl.), **hot-path truth**, **memory & resource leaks**, **I/O & wire budgets**, **measured scaling**. Any language/project: discover repo's own benchmark/profiler/load/query-log/browser-perf tooling; never assume stack. Weighs every cost in context (Guardrails). Not general code review (use `/code-review`); no inline auto-fixes — engineer owns fix. Opt-in remediation on request: one solver per blocker, re-measured before merge (step 11). On invoke: state measurement plan; fixes opt-in — per-blocker solvers, re-measured before merge. Then gather inputs. Whatever the target's phrasing — a change to check, or a question about perf-review itself — it's the surface to measure; never substitute a hand-written explanation for actually running the gates below. ## Operating loop 1. **Ground the target and baseline.** First discover project's own perf tooling + invocation (`CLAUDE.md`/`AGENTS.md`, README, CI config, dependency manifest): bench runner, profiler, load generator, query log, (UI) browser-perf. Examples — bench: `pytest-benchmark`, `go test -bench`, `cargo bench`/`criterion`, JMH, `hyperfine`, own suite; profiler: `cProfile`/`py-spy`, `pprof`, `perf`, `async-profiler`, Instruments, browser devtools; load/soak: `wrk`, `k6`, `locust`, `ab`; memory/leak: `tracemalloc`, `memray`, `valgrind`, `heaptrack`, `/proc` + fd counts. Page/fr