internal-investigation
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted external data (document batches, emails, and interview notes) which are subsequently used for high-stakes tasks like querying the investigation history and drafting legal memorandums.
- Ingestion points: In
SKILL.md(Mode 2: Add data), the skill is designed to ingest "document batches", "emails", "interview notes", and "files" provided by the user. - Boundary markers: While the skill utilizes a structured YAML format (
log.yaml) for data storage, it lacks explicit delimiters or instructions to the model to treat the content of thesummaryorquotefields as non-executable text, potentially allowing embedded instructions to be followed during later retrieval steps. - Capability inventory: The skill possesses file-write capabilities across the investigation workspace (e.g.,
memo.md,log.yaml,sources-checklist.yaml) and provides the agent with the ability to summarize and synthesize the entire investigation history. - Sanitization: There is no evidence of sanitization, filtering, or escaping of the ingested external content before it is interpolated into the persistent log files or investigation drafts.
Audit Metadata