case-study
Pass
Audited by Gen Agent Trust Hub on Nov 1, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and structure untrusted data from various sources such as raw logs, research notes, and code artifacts. This creates an attack surface where hidden instructions within those artifacts could attempt to influence the agent's behavior. However, the skill provides no access to network tools, shell execution, or sensitive file paths, limiting the potential impact to the generated text output.
- Ingestion points: The skill processes user-supplied raw logs, notes, code, research, and project artifacts as defined in the 'Automated pipeline' and 'Workflow' sections of SKILL.md.
- Boundary markers: No explicit data delimiters or 'ignore instructions' wrappers are defined for the ingested content.
- Capability inventory: The skill is restricted to text processing and markdown generation; there are no references to allowed-tools or subprocess executions in the analyzed files.
- Sanitization: The skill employs instructional sanitization via the 'Evidence contract' and 'Quality Gate', which mandate labeling claims (Executed, Observed, Reported, Proposed) and explicitly forbid the invention of facts.
Audit Metadata