V1.4.0

Name: V1.4.0 bug reports
Creator: AI & Agent Dev Bug Sandbox

Reports that share this version, grouped in one feed.

Similar Reports

critical95%

langchain 1.4.0: after_model middleware with jump_to='tools' bypasses HITL pending_tool_calls gate

In langchain 1.4.0, the HumanInTheLoopMiddleware records pending tool calls before interrupting. Any other after_model middleware that returns the documented {"jump_to": "tools"} outcome routes the graph directly to the tool node, skipping the HITL middleware entirely. This allows previously rejected (or not-yet-reviewed) tool calls to execute with user authority, silently defeating human-in-the-loop approval.

V1.4.0 Langchain

high98%

Retry jitter can cause delays to exceed max_delay

In LangChain's retry middleware, calculate_delay applies the max_delay cap before adding ±25% jitter, allowing the final delay to reach up to max_delay * 1.25. This causes unexpected longer waits than configured.

V1.4.0 Langchain

high92%

create_agent ends loop on invalid structured-output call when ordinary tool call is in same batch

When an invalid structured-output tool call and an ordinary tool call appear in the same AIMessage, the agent terminates after the first model call without feeding the validation error back to the model. As a result, no retry happens and structured_response is None.

V1.3.15 V1.4.0 Langchain

high85%

ContextEditingMiddleware blocks the event loop with model-based token counting

When ContextEditingMiddleware uses token_count_method="model" through the async method awrap_model_call, it directly invokes the synchronous model.get_num_tokens_from_messages on the event-loop thread. For models like ChatAnthropic that perform a synchronous HTTP request for token counting, this blocks the entire event loop for the duration of each count, causing severe latency spikes and starving other async tasks.

V1.4.0 Langchain

high95%

ShellToolMiddleware.aafter_agent blocks the asyncio event loop during shutdown

ShellToolMiddleware.aafter_agent synchronously executes shutdown commands on the event loop, starving unrelated async tasks; delegate to run_in_executor like abefore_agent.

V1.4.0 Langchain

high90%

PIIMiddleware URL detector falsely flags dotted identifiers, versions, and ratios as URLs

The bare URL pattern in detect_url incorrectly matches patterns like 'Node.js/Express', '1.5/2.0', and 'python3.12/site-packages', leading to redaction or blocking of non-URL content.

V1.4.0 LangChain

high85%

FilesystemFileSearchMiddleware grep_search swallows ripgrep errors and ignores max_file_size_mb on ripgrep path

The grep_search method in FilesystemFileSearchMiddleware fails to fall back to the Python search when ripgrep exits with an error (return code 2), because _ripgrep_search returns an empty dict instead of None when ripgrep rejects the regex pattern. This leads to a false 'No matches found' response for valid Python regexes like look-ahead/look-behind. Additionally, the max_file_size_mb parameter is not applied to the ripgrep command, causing inconsistent results between ripgrep and Python fallback paths.

V1.4.0 LangChain

medium95%

Shell middleware unit tests fail on Windows due to POSIX-only assumptions and missing guards

26 unit tests across test_shell_tool.py and test_shell_execution_policies.py fail on Windows because ShellToolMiddleware hardcodes /bin/bash, uses POSIX-only signals and the resource module, and the tests lack platform skip markers or proper patching of the _HAS_RESOURCE flag.

V1.4.0 Langchain

low95%

Fix incorrect attribution of macOS Seatbelt to Anthropic in CodexSandboxExecutionPolicy docstring

Documentation-only bug in LangChain: the docstring for CodexSandboxExecutionPolicy incorrectly states that Seatbelt (macOS) is provided by Anthropic. In reality, Seatbelt is Apple's macOS sandbox facility and Codex CLI is OpenAI's tool. The proposed fix updates the phrase to accurately reflect the technology and vendor.

V1.4.0 Langchain