ingest
Pass
Audited by Gen Agent Trust Hub on Aug 10, 2026
Risk Level: SAFEPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by ingesting and processing external data.
- Ingestion points: The agent fetches content from user-specified URLs and reads files from the `0
- Inbox/
folder (SKILL.md`). - Boundary markers: The instructions do not define clear delimiters or specific safety instructions to isolate the fetched content from the agent's core logic.
- Capability inventory: The agent can read, write, and move files using the
obsidiantoolset, and execute shell commands for formatting dates. - Sanitization: No input sanitization or validation is performed on the ingested content before synthesis.
- [DATA_EXFILTRATION]: The skill performs network operations to external, non-whitelisted domains.
- Network Activity: The skill instructions explicitly direct the agent to "fetch the content directly" for URLs or web content provided by the user (
SKILL.md). This constitutes network access to arbitrary external domains.
Audit Metadata