semantic-entity-topical-map-audit
Pass
Audited by Gen Agent Trust Hub on Sep 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external websites specified by the user to perform entity and topical audits.
- Ingestion points: Website HTML and JSON-LD data are fetched via
curlinreferences/checks.md(Checks 1, 3, 4, 5, 6). - Boundary markers: The instructions do not define explicit boundary markers or instructions for the agent to ignore potentially malicious content embedded in the target website's metadata.
- Capability inventory: The skill processes fetched data using Python scripts to extract entity types, IDs, and link structures, which are then summarized for the user.
- Sanitization: The skill uses
json.loadsfor structured data and regular expressions for extraction, which offers some structural protection, but the semantic content remains unvalidated. - [COMMAND_EXECUTION]: The skill utilizes shell commands (
curl,grep,python3) to interact with target websites and Wikidata. - Evidence:
references/checks.mdcontains several automated checks that execute shell pipelines to process website content. - Context: The operations are limited to standard web auditing tasks and data retrieval from a well-known service (Wikidata).
Audit Metadata