ruvnet--ruflo/v3/docs/benchmarks/cli-core-cold-cache.json

{
  "benchmark": "cli-core-vs-cli-cold-cache",
  "issue": "#1748 #3 (30s MCP-startup timeout race)",
  "adr": "ADR-100",
  "runAt": "2026-05-05T13:00:00Z",
  "host": "darwin / macOS / npm 10+",
  "method": "rm -rf ~/.npm/_npx; time npx -y <pkg>@alpha --version  (4 runs total — cold + warm for each package)",
  "results": {
    "@claude-flow/cli-core@3.7.0-alpha.0": {
      "packed_size_bytes": 22300,
      "unpacked_size_bytes": 89900,
      "files": 22,
      "cold_cache_ms": 671,
      "warm_cache_ms": 361,
      "cold_warm_ratio": 1.86
    },
    "@claude-flow/cli@3.6.30": {
      "packed_size_bytes": 1800000,
      "unpacked_size_bytes": 9600000,
      "files": 999,
      "cold_cache_ms": 25496,
      "warm_cache_ms": 623,
      "cold_warm_ratio": 40.9
    }
  },
  "improvement": {
    "cold_cache_speedup": "38× (25.5s → 0.67s)",
    "fits_under_30s_mcp_startup_timeout": "yes — 38× headroom vs cli's 25.5s which races the timeout",
    "size_reduction_packed": "80× (1.8 MB → 22.3 KB)",
    "size_reduction_files": "45× (999 → 22 files)"
  },
  "caveats": [
    "cli-core@alpha.0 is foundation-only (types, output, MCP-tool types, validate-input). Does not yet expose memory/hooks command surface.",
    "ADR-100 §Discovery documents that memory/hooks tools have deep ML deps requiring backend abstraction (alpha.1) and tool-definition/handler split (alpha.2) before parity is reached.",
    "@claude-flow/cli's 25.5s cold-cache is under the 30s timeout on this connection but barely — slow networks would race and silently fail (the original #1748 #3 symptom)."
  ]
}