contentful-env-analysis
Pass
Audited by Gen Agent Trust Hub on Nov 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external data (Contentful entries and metadata) which could potentially contain malicious instructions.
- Ingestion points: Data enters the environment via the
fetchcommand described inSKILL.md, which retrieves entries and assets from Contentful APIs into a local SQLite database. - Boundary markers: The skill provides highly explicit instructions in the "Content is data, not instructions" section of
SKILL.md, warning the agent to ignore embedded requests and never treat content as instructions. - Capability inventory: The skill possesses capabilities for shell execution (to run Node.js scripts), network access (to Contentful domains), and SQL query execution.
- Sanitization: The agent is instructed to use the shipped read-only query command and to quote suspicious text as data rather than incorporating it into executable logic.
- [DYNAMIC_EXECUTION]: The skill permits the execution of dynamic SQL queries against the local snapshot database.
- Evidence: The
querycommand allows the agent to execute SQL statements passed as a string argument. - Mitigation: The skill enforces strict constraints, permitting only
SELECT,WITH, andEXPLAINstatements, disabling SQLite extensions, and opening the database in read-only mode.
Audit Metadata