microbench

Overview ¶

microbench aggregates per-trial JSON files into a head-to-head report (ghostchrome vs playwright-mcp) and writes:

• benchmark/results.md (human-readable table)
• benchmark/results.json (machine, for tooling)
• benchmark/badges/{tokens,latency,size}.json (shields.io endpoint)

Input JSON shape (one trial per file in --input dir):

{"tool":"ghostchrome|playwright-mcp", "url":"...", "site":"<label>",
 "bytes":1234, "durationMs":567, "peakRssKb":89000}