utopia
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill manages a bitemporal knowledge application that ingests external documents and provides search/chat capabilities, creating a surface where malicious instructions embedded in processed data could influence the agent.
- Ingestion points: Ingests external documents and knowledge base data for search, chat, and derivation tasks as defined in
SKILL.md. - Boundary markers: Instructions mandate human adjudication for uncertain facts and agent-proposed write operations to maintain a "review boundary" (specified in
SKILL.md). - Capability inventory: The agent has access to
Bash,Read,Grep, andGlobtools for file and system interactions (SKILL.md). - Sanitization: While a review boundary is established, specific automated sanitization or escaping of ingested document content before model processing is not detailed.
- [COMMAND_EXECUTION]: The skill uses the
Bashtool to perform application management tasks, such as deployment and backups. - Evidence:
SKILL.mdandreferences/upstream.mdmandate explicit approval gates for all destructive actions, database mutations, or service deployments. - Mitigation: The provided
scripts/preflight.pyhelper facilitates safe, read-only integrity checks of the local environment using secure POSIX path traversal and hash verification before any operational commands are executed.
Audit Metadata