abhigyanpatwari/GitNexus · error · SandboxError

transcript artifact exceeds the bounded run-output limit

Error message

transcript artifact exceeds the bounded run-output limit

What it means

After type checks, the size must satisfy 0 <= bytes <= runner.MAX_TRANSCRIPT_BYTES (8 MiB). A negative value or a value above the bound is rejected so a single oversized (or crafted) transcript cannot exhaust proposer context or memory during evidence read.

Source

Thrown at eval/workflow_bench/evolve.py:340

    return root / Path(*relative.parts)


def _transcript_artifact_metadata(metadata: Any) -> tuple[str, str, int]:
    """Validate transcript metadata without touching any host path."""

    if not isinstance(metadata, dict) or set(metadata) != {"path", "sha256", "bytes", "source"}:
        raise SandboxError("transcript artifact metadata must contain only path, sha256, bytes, and source")
    relative = metadata["path"]
    expected_digest = metadata["sha256"]
    expected_size = metadata["bytes"]
    if metadata["source"] != runner_sessions.PARENT_EVENT_STREAM_SOURCE:
        raise SandboxError("transcript artifact source is not the parent event stream")
    if not isinstance(relative, str) or not re.fullmatch(r"[0-9a-f]{64}", str(expected_digest)):
        raise SandboxError("transcript artifact metadata is malformed")
    if not isinstance(expected_size, int) or isinstance(expected_size, bool):
        raise SandboxError("transcript artifact byte count must be an integer")
    if expected_size < 0 or expected_size > runner.MAX_TRANSCRIPT_BYTES:
        raise SandboxError("transcript artifact exceeds the bounded run-output limit")
    return relative, expected_digest, expected_size


def _normalized_transcript_artifact_path(relative_value: str) -> str:
    """Apply the transcript path contract without touching the filesystem."""

    relative = PurePosixPath(relative_value)
    if (
        relative.is_absolute()
        or len(relative.parts) != 2
        or relative.parts[0] != "transcripts"
        or any(part in {"", ".", ".."} for part in relative.parts)
    ):
        raise SandboxError(f"unsafe results artifact path: {relative_value!r}")
    return relative.as_posix()


def _preflight_transcript_artifacts(evidence: list[dict[str, Any]]) -> list[list[Any]]:

View on GitHub (pinned to d540b00184)

Solutions

1. Reduce transcript verbosity at capture time (drop redundant tool outputs) so the file stays under 8 MiB.
2. If a genuinely large transcript is needed, raise runner.MAX_TRANSCRIPT_BYTES deliberately and audit the proposer context budget.
3. Re-measure the size after the file is fully written and rewrite the row.

Example fix

# before: 12 MiB transcript admitted
row = {..., "bytes": 12 * 1024 * 1024}

# after: trim at capture time, then record the real post-trim size
size = path.stat().st_size
assert size <= runner.MAX_TRANSCRIPT_BYTES, size
row = {..., "bytes": size}

from eval.workflow_bench.runner_sessions import MAX_TRANSCRIPT_BYTES

def transcript_size_within_bound(metadata: dict) -> bool:
    n = metadata.get("bytes")
    return isinstance(n, int) and not isinstance(n, bool) and 0 <= n <= MAX_TRANSCRIPT_BYTES

When it happens

Trigger: A transcript larger than 8 MiB (verbose agent session with full event echo); a negative size from a corrupt row; a size computed before the file finished writing.

Common situations: Very long sessions; debug-level event capture; a race where the size was read mid-write; integer underflow in a custom writer.

Related errors

• {label} exceeds the bounded evidence limit
• transcript artifact metadata must contain only path, sha256,
• transcript artifact source is not the parent event stream
• transcript artifact metadata is malformed
• transcript artifact byte count must be an integer

AI-assisted analysis of abhigyanpatwari/GitNexus@d540b00184 (2026-08-12). Data as JSON: /api/errors/eb2d107bfe47ee42. Report an issue: GitHub.