fix(langchain): backport patch ReDoS vulnerability in MRKL and ReAct action regex (CVE-2024-58340)

[Eugene Yurtsev (eyurtsev)]

Backport of #35598 to the v0.3 branch. Patches the ReDoS vulnerability in MRKL and ReAct agent action regex patterns (CVE-2024-58340).

Created with Deep Agents CLI.

[Copilot]

Pull request overview

This is a backport of PR #35598 to the v0.3 branch, patching a ReDoS (Regular Expression Denial of Service) vulnerability (CVE-2024-58340) in the MRKL and ReAct agent action regex parsers. The vulnerable pattern (.*?)[\s]*Action allowed catastrophic backtracking on crafted input because (.*?) with re.DOTALL and [\s]* both compete to match whitespace characters, creating exponential ambiguity.

Changes:

• Remove the redundant [\s]* quantifier between the first capture group and the literal Action keyword in both MRKLOutputParser.parse() and ReActSingleInputOutputParser.parse().
• Add SIGALRM-based regression tests for both parsers to verify regex completes in bounded time on adversarial input.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated no comments.

File	Description
libs/langchain/langchain/agents/mrkl/output_parser.py	Remove [\s]* from action-parsing regex to fix ReDoS vulnerability
libs/langchain/langchain/agents/output_parsers/react_single_input.py	Same regex fix applied to the ReAct single input parser
libs/langchain/tests/unit_tests/agents/test_mrkl_output_parser.py	Add ReDoS regression test for MRKLOutputParser
libs/langchain/tests/unit_tests/agents/output_parsers/test_react_single_input.py	Add ReDoS regression test for ReActSingleInputOutputParser

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

You can also share your feedback on Copilot code review. Take the survey.