| """Fixed zero-action mailbox v2 client; infrastructure check, never a model score. |
| |
| This process reads only its current mailbox request and referenced RGB images. |
| It does not import the harness, native simulator, task catalog or scorer. |
| """ |
|
|
| import argparse |
| import hashlib |
| import io |
| import json |
| import os |
| from pathlib import Path |
| import time |
|
|
|
|
| def public_path(root, relative): |
| path = Path(relative) |
| if path.is_absolute() or ".." in path.parts: |
| raise ValueError("v2 path must be mailbox-root relative") |
| resolved = (root / path).resolve() |
| if root not in resolved.parents: |
| raise ValueError("path escapes mailbox root") |
| return resolved |
|
|
|
|
| def main(): |
| parser = argparse.ArgumentParser(description=__doc__) |
| parser.add_argument("--directory", type=Path, required=True) |
| parser.add_argument("--decisions", type=int, default=2) |
| parser.add_argument("--chunk-size", type=int, default=1) |
| parser.add_argument("--timeout-s", type=float, default=300) |
| args = parser.parse_args() |
| if args.decisions < 1 or not 1 <= args.chunk_size <= 5 or not 0 < args.timeout_s <= 3600: |
| parser.error("positive decisions, chunk-size 1..5 and timeout 0..3600 required") |
| from PIL import Image |
|
|
| root = args.directory.resolve() |
| deadline = time.monotonic() + args.timeout_s |
| seen = set() |
| images_checked = 0 |
| while time.monotonic() < deadline: |
| ready_path = root / "ready.json" |
| if not ready_path.exists(): |
| time.sleep(0.05) |
| continue |
| ready = json.loads(ready_path.read_text()) |
| if ready["schema_version"] != "starvla-rsi.mailbox/v2": |
| raise ValueError("this diagnostic only supports mailbox v2") |
| if ready["status"] == "failed": |
| raise RuntimeError("evaluator reported failed decision") |
| if ready["status"] == "episode_closed": |
| print(json.dumps({"kind": "fixed_zero_action_diagnostic_not_model_score", |
| "submitted_decisions": len(seen), "images_checked": images_checked, |
| "provider_calls": 0, "native_success": None})) |
| return |
| if ready["status"] != "pending" or ready["request_id"] in seen: |
| time.sleep(0.05) |
| continue |
| if len(seen) >= args.decisions: |
| raise RuntimeError("evaluator requested more than declared diagnostic decisions") |
| observation = json.loads(public_path(root, ready["observation_path"]).read_text()) |
| if observation["request_id"] != ready["request_id"]: |
| raise ValueError("request/observation mismatch") |
| budgets = observation["budgets"] |
| if (budgets["max_policy_calls"] != args.decisions |
| or budgets["replan_interval"] != args.chunk_size |
| or budgets["primary_budget"] != "policy_decisions"): |
| raise ValueError("diagnostic requires matching decision-primary budgets") |
| for frame in observation["images"].values(): |
| data = public_path(root, frame["path"]).read_bytes() |
| if hashlib.sha256(data).hexdigest() != frame["sha256"]: |
| raise ValueError("RGB hash mismatch") |
| with Image.open(io.BytesIO(data)) as image: |
| image.load() |
| if image.mode != "RGB": |
| raise ValueError("diagnostic expects RGB PNG") |
| images_checked += 1 |
| response = public_path(root, ready["response_path"]) |
| temporary = response.with_suffix(".diagnostic-tmp") |
| with temporary.open("x") as stream: |
| json.dump({"request_id": ready["request_id"], |
| "actions": [[0.0] * 7 for _ in range(args.chunk_size)], |
| "usage": {"calls": 0, "input_tokens": 0, "output_tokens": 0, |
| "cached_input_tokens": 0, "estimated_cost_usd": 0}, |
| "note": "Fixed zero-action infrastructure diagnostic, no model."}, stream) |
| os.replace(temporary, response) |
| seen.add(ready["request_id"]) |
| raise TimeoutError("diagnostic deadline exceeded") |
|
|
|
|
| if __name__ == "__main__": |
| main() |
|
|
