ln-840-benchmark-compare

> **Paths:** File paths (`shared/`, `references/`) are relative to skills repo root. Locate this SKILL.md directory and go up one level for repo root. # Benchmark Compare **Type:** L3 Worker **Category:** 8XX Optimization -> 840 Benchmark Run a clean A/B benchmark in Claude Code: one session with built-in tools only, one with `hex-line`. The benchmark is scenario-based, diff-validated, manifest-driven, and runtime-backed. It measures activation, correctness, time, cost, and tokens. The current runner is intentionally scoped to this internal A/B. It does not, by itself, prove best-in-class against external alternatives. --- ## Input / Output | Direction | Content | |-----------|----------| | **Input** | Repo checkout containing `mcp/hex-line-mcp/`, optional `references/goals.md`, optional `references/expectations.json` | | **Output** | Comparison report in `skills-catalog/ln-840-benchmark-compare/results/{date}-comparison.md` plus machine-readable benchmark summary artifact | --- ## Prerequisites - `claude --version` succeeds - `git` succeeds - `mcp/hex-line-mcp/server.mjs` exists - `mcp/hex-line-mcp/hook.mjs` exists - `skills-catalog/ln-840-benchmark-compare/references/goals.md` exists - `skills-catalog/ln-840-benchmark-compare/references/expectations.json` exists - `skills-catalog/ln-840-benchmark-compare/references/mcp-bench.json` exists --- ## Quick Run ```bash bash skills-catalog/ln-840-benchmark-compare/scripts/run-benchmark.sh \ [skills-catalog/ln-840-benc...