citation-graph-ingest
Pass
Audited by Gen Agent Trust Hub on Aug 13, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill uses native platform CLI tools (gbrain) for all operations including searching, reading, and linking documents. It does not require external network access or access to sensitive local file paths.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted third-party documents which presents an indirect prompt injection surface. This is acknowledged and mitigated in the skill instructions.
- Ingestion points: Document content is retrieved via gbrain search and get tools (SKILL.md).
- Boundary markers: The skill includes explicit instructions for the agent to treat document text strictly as DATA and to ignore any embedded imperatives (SKILL.md).
- Capability inventory: The skill has the capability to write typed edges to a graph database via gbrain link (SKILL.md).
- Sanitization: The process relies on model judgment to classify relationships based on context while ignoring directives within the text.
- [SAFE]: The deterministic hit for PI_CONCEALMENT is a false positive. The skill transparently explains its durable write behavior and idempotent skipping of duplicates, which are standard technical operations and do not represent attempts to hide malicious activities from the user.
Audit Metadata