langchain 1.4.6

Source: https://github.com/langchain-ai/langchain
Tag scanned: langchain==1.4.6
Directory: libs/langchain_v1 (only this part of the repository was scanned)
Compared with: langchain==1.4.5 (0 signals introduced since)
Files: 166 scanned, 24 skipped (binary, oversized, or unrecognised)
Signals: 0 critical · 0 warn · 1 info
Scan time: 3.7 s
LLM review: Reviewed by claude-haiku-5-5 (Anthropic): 1 of 1 warn and critical signal; 1 demoted to info.
Disclude version: 4.0.0
Scanned (UTC): 2026-10-15 02:25

Signals

Ordered by severity, then signals introduced in this release. Hidden characters are shown as ⟨U+…⟩.

info ~~warn~~ → info by review dynamic-execution libs/langchain_v1/tests/unit_tests/agents/middleware/implementations/test_tool_emulator.py:47:23

`eval` called on a non-literal expression (ast pass, confidence 0.70)
```
    # This tool executes normally (not emulated)
    return f"Result: {eval(expression)}"  # noqa: S307
```
likely benign LLM, confidence 0.85 A test fixture tool that evaluates an expression with eval to simulate a non-emulated tool result.
Reasoning
The eval call sits in a unit test file for a tool emulator middleware. The surrounding comment says the tool 'executes normally (not emulated)', which indicates a test fixture that stands in for a calculator-style tool. The `# noqa: S307` marker shows the author knowingly accepted the eval use, as is common in test code. Nothing in the snippet is obfuscated, encoded, or tied to a dropper or network activity. The residual doubt is that eval on untrusted input is risky in general, but here the input comes from test code.

Reproduce This Scan

Run disclude locally to verify. Add --format sarif for SARIF output. Add --llm (with an API key in the environment) to include the LLM review.

cargo install disclude --version 4.0.0
git clone https://github.com/langchain-ai/langchain langchain && cd langchain
git checkout langchain==1.4.6
disclude scan --severity info --diff langchain==1.4.5 libs/langchain_v1