feed-operations
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill documents procedures for bulk data ingestion from external sources, creating an attack surface for indirect prompt injection if the ingested content is subsequently processed by an AI agent.
- Ingestion points:
docs/feed-clients.mdincludes examples of fetching and piping data from external URLs into thevespa feedcommand. - Boundary markers: The documentation does not provide specific delimiters or instructions for the agent to isolate or ignore potentially malicious instructions within the data being fed.
- Capability inventory: The skill enables the agent to execute shell commands using the
vespaCLI andcurlfor interacting with the database. - Sanitization: The documentation does not specify methods for sanitizing the ingested document data.
- [EXTERNAL_DOWNLOADS]: The documentation contains examples illustrating how to fetch data from remote endpoints.
- Evidence:
docs/feed-clients.mddemonstrates fetching data fromhttps://example.com/data.jsonlusingcurl. - [COMMAND_EXECUTION]: The skill provides instructions for the agent to execute shell commands to manage documents in a Vespa cluster.
- Evidence:
SKILL.mdanddocs/feed-clients.mdlist numerous commands using thevespaCLI (e.g.,vespa document put,vespa feed) andcurlfor standard REST operations.
Audit Metadata