legal-data-hunter-pl
Warn
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATIONDYNAMIC_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill clones an external code repository from a well-known service (https://github.com/worldwidelaw/legal-sources.git) to the local environment.
- [REMOTE_CODE_EXECUTION]: The skill involves executing logic from the downloaded external repository using python runner.py, which runs code that is not part of the skill's own distribution.
- [COMMAND_EXECUTION]: The skill uses Bash tools to perform repository management, git sparse-cloning, and execution of the data harvesting framework.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted legal data (legislative acts, court judgments, regulator decisions) from various government portals. 1. Ingestion points: runner.py fetches documents from sources like SAOS, KNF, and various court APIs (SKILL.md, references/coverage-pl.md). 2. Boundary markers: The skill does not explicitly define prompt delimiters or ignore-embedded-instruction warnings for the fetched legal text before processing. 3. Capability inventory: The skill has access to Bash, WebFetch, Read, and Write tools (SKILL.md). 4. Sanitization: While content is normalized to a common schema, there is no mention of specific prompt sanitization or instruction filtering for the ingested document text.
- [DATA_EXFILTRATION]: The skill author warns that the common/pdf_extract.py script in the Legal Data Hunter project queries a third-party Neon Postgres database by default, which presents a risk of metadata or sensitive information being transmitted to an external service.
- [DYNAMIC_EXECUTION]: The skill relies on executing downloaded Python scripts and mentions dependencies like playwright for headless browser operations, which can involve the runtime execution of browser-based logic and external scripts.
Audit Metadata