eu-sparql-search

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 1.00). The skill contains explicit instructions to bypass platform/web_fetch restrictions (use bash_tool to "work unconditionally") and to disable SSL verification, which are deceptive and outside the declared safe usage/allowed-tools of the skill.

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). In SKILL.md, the workflow executes user-driven SPARQL (via bash_tool with Python/urllib) against https://publications.europa.eu/webapi/rdf/sparql, then can fetch and HTML-extract full document content from item URLs returned by that query; this ingests free-text supplied at runtime (e.g., the user’s SPARQL/query intent) to determine what external content to read.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 0.80). The skill explicitly instructs the agent to bypass platform permission checks by using bash_tool instead of web_fetch and even suggests disabling SSL verification, which encourages circumventing security mechanisms on the host environment.