dify-docs-reader-test
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill reads external documentation files and persona descriptions to provide context for a subagent review process, creating a potential surface for indirect prompt injection.
- Ingestion points: Local files identified by
{PATHS}and reader personas provided via{PERSONA}in the dispatch prompt (SKILL.md). - Boundary markers: The dispatch template lacks explicit delimiters or structural boundaries for the injected content, though it provides specific instructional constraints to the subagent.
- Capability inventory: The skill employs the
Agenttool to spawn subagents capable of reading filesystem content. - Sanitization: The skill does not perform sanitization or escaping of the content read from the target files.
- [COMMAND_EXECUTION]: The skill automates the dispatch of general-purpose subagents to perform analysis tasks. This utilizes the platform's native agent capabilities to process file content, which is consistent with the skill's stated documentation review purpose.
Audit Metadata