benchforgelisted

# Universal Benchmarker Skill (`BenchForge`) `BenchForge` is an **Open Scientific Evidence Infrastructure for the Agentic Era**. It provides a protocol-driven framework for defining, validating, executing, and documenting benchmarks across AI Skills, Software Systems, Hardware, and Workflows. > **Full Protocol Specification & MVP Plan**: > Refer to the complete specification at [docs/guide.md](file:///D:/Liem%20Product/Liem%20Skills/skills/benchforge/docs/guide.md). --- ## 1. User Operating & Usage Guide ### How to Trigger the Skill An AI Agent or user can activate `BenchForge` using natural language triggers or skill invocations: - *"Benchmark `<candidate-skill>` vs `<baseline-prompt>`"* - *"Bandingkan performa `<candidate-tool>` vs `<baseline-tool>` secara data-oriented"* - *"Jalankan benchmark suite untuk proyek ini di folder `./benchmarks/`"* - *"Generate `BENCHMARK_SUMMARY.md` dan `BENCHMARK.md` untuk GitHub"* --- ### Step-by-Step Automated Execution Lifecycle When invoked, the agent executes the following 4-step workflow: ```mermaid sequenceDiagram autonumber actor User as User / Developer participant BDL as BDL Spec Parser (.bench.yaml) participant Harness as Execution Sandbox participant Stat as Dual Statistician Engine participant Report as Dual Report Generator User->>BDL: "Benchmark <candidate-skill> vs <baseline-prompt>" BDL->>BDL: 1. Generate BDL spec & multi-variable dataset BDL->>Harness: 2. Execute N-run workloa