113-java-maven-documentation
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill interacts with external
pom.xmlfiles which could contain untrusted data. It incorporates specific safety constraints to minimize the risk of indirect prompt injection. - Ingestion points: The skill queries the root and submodule
pom.xmlfiles using local tooling (Step 1 and Step 2 inSKILL.md). - Boundary markers: The instructions explicitly label these files as 'UNTRUSTED POM INPUT' and direct the agent not to load full files or free-text fields like descriptions and comments into its context.
- Capability inventory: The skill's primary capability is generating a local documentation file (
DEVELOPER.md). It does not perform network operations or execute untrusted code. - Sanitization: The workflow requires 'ALLOWLISTED EXTRACTION', where only structural fields (artifact IDs, module paths, plugin names) are processed, effectively filtering out potential injection vectors in free-text fields.
Audit Metadata