legal-data-hunter-pl

Fail

Audited by Gen Agent Trust Hub on Jul 14, 2026

Risk Level: HIGHREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill instructions direct the agent to clone an external third-party repository (https://github.com/worldwidelaw/legal-sources.git) and run its scripts (runner.py) using Python. Executing code from an unverified external source is a high-risk activity that can lead to arbitrary command execution on the host system.\n- [EXTERNAL_DOWNLOADS]: The skill relies on fetching core framework code and legal data collectors from a non-trusted external GitHub repository. This introduces a supply chain risk where the remote code could be modified to include malicious behavior.\n- [DATA_EXFILTRATION]: The skill documentation notes that internal components (common/pdf_extract.py) and search functionalities interact with a remote database hosted on Neon and a hosted API (legaldatahunter.com). While the skill provides explicit warnings to users to avoid these services to protect data privacy, the presence of these network-dependent features in the underlying harvesting logic creates a surface for data leakage to third-party infrastructure.\n- [COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to perform repository management tasks (git clone, sparse-checkout) and to invoke the Python harvesting scripts. This automated execution of shell commands, especially when combined with external code, increases the risk of system compromise.\n- [PROMPT_INJECTION]: The skill is designed to ingest and process large quantities of unstructured legal data from various government portals, creating an indirect prompt injection surface:\n
  • Ingestion points: Fetches documents and API responses from official Polish legal portals (e.g., SAOS, Sejm ELI) via collectors in sources/PL/.\n
  • Boundary markers: No explicit markers are used to isolate the harvested text or instruct the agent to ignore instructions embedded within the data.\n
  • Capability inventory: The skill has access to Bash (scripting), Write (file system access), and WebFetch (network operations).\n
  • Sanitization: There is no documentation of sanitization processes to filter out malicious content or instructions within the retrieved legal documents.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Jul 14, 2026, 08:28 AM
Security Audit — agent-trust-hub — legal-data-hunter-pl