langchain-attack-probe
Pass
Audited by Gen Agent Trust Hub on Jun 13, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: While the skill contains instructions like 'Ignore previous instructions', these are examples of test payloads (e.g., LC-RAG-001) used to verify if a target application is vulnerable to indirect prompt injection. They are not intended to hijack the agent analyzing or running the skill itself.
- [REMOTE_CODE_EXECUTION]: The skill mentions and provides examples of Python REPL and Shell tools. However, it does not execute these commands itself; rather, it documents how their misuse in LangChain applications leads to RCE and provides payloads to test for such vulnerabilities (e.g., LC-REPL-001, LC-SHELL-001).
- [DATA_EXFILTRATION]: The skill discusses SSRF (LC-SSRF-001) and callback leakage (LC-CB-001) as potential risks in misconfigured LangChain apps. It does not contain code to exfiltrate the user's data; it provides methods to test if a target app might leak metadata or logs.
- [COMMAND_EXECUTION]: The skill contains shell and python snippets, but these are clearly marked as educational examples ('Wrong vs. Right') or as testing payloads for the user to manually apply to their own infrastructure.
- [TRUSTED_SOURCE]: The skill references documentation and repositories from 'langchain-ai', 'langchain', and 'python.langchain.com', which are well-known and trusted entities in the AI development ecosystem.
Audit Metadata