Dense, information-rich body of tables, decision trees, and pointers that assumes Claude's competence; avoids explaining concepts Claude already knows (no "what is pprof" padding).

Provides concrete executable commands and thresholds — `go test -bench=... -benchmem -count=6`, `benchstat report-1.txt report-2.txt`, "GOMEMLIMIT to 80-90%", "MaxIdleConnsPerHost defaults to 2" — and maps pprof signals to specific actions.

The iterative cycle is an explicitly numbered 8-step sequence with a validation checkpoint (benchstat compare for statistical significance), a commit step, and a repeat/feedback loop.

Clear overview routing to one-level-deep reference files (memory.md, cpu.md, io-networking.md, runtime.md, caching.md, observability.md), all of which exist in ./references/, with cross-references to sibling skills.

Lists multiple concrete optimization actions — "allocation reduction, CPU efficiency, memory layout, GC tuning, pooling, caching, and hot-path optimization" — matching the multi-action anchor.

Explicitly answers both what (Golang performance optimization patterns) and when ("Use when profiling or benchmarks have identified a bottleneck"; "Also use when performing performance code review"), with explicit triggers.

Natural trigger terms users would say are well covered: "profiling", "benchmarks", "bottleneck", "performance code review", "optimization pattern".

Clear Go-performance niche with explicit conflict avoidance via "Not for measurement methodology" and "Not for ... debugging workflow" redirecting to sibling skills.