ddd4j-quarkus-release
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The file
references/source-evidence.mdcontains a hardcoded absolute directory path (/Users/wandl/workspaces/workspace-ddd4j/workspace-ddd4j-boot/ddd4j-quarkus). Hardcoding specific developer home directories exposes information about the source environment and may lead to information leakage or operational errors if the agent attempts to resolve these paths on different systems.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted metadata and project source files, establishing an attack surface for indirect prompt injection.\n - Ingestion points: Project configuration (POM), source code, and CI evidence logs are ingested as part of the release validation process (referenced in
references/source-evidence.md).\n - Boundary markers: The skill lacks explicit delimited structures to differentiate untrusted external content from the agent's internal instruction sequence.\n
- Capability inventory: The skill is capable of executing Maven build cycles, interacting with Git repositories, and performing authenticated artifact publishing to private servers.\n
- Sanitization: The instructions mandate the removal and redaction of plain-text credentials, tokens, and OIDC secrets from all reports and logs, which reduces data exfiltration risks but does not address malicious instruction influence.
Audit Metadata