fabric-workflow
Pass
Audited by Gen Agent Trust Hub on Jul 25, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface where external data is interpolated into instructions for sub-agents.\n
- Ingestion points: The variable π.task (user-controlled objective) is used in the prompts for the Discovery, Analyze, and Verify phases. Data retrieved from the file system (inventory items) and generated findings are also interpolated into subsequent agent prompts.\n
- Boundary markers: There are no boundary markers or explicit delimiters used to isolate untrusted user input or file data from the agent's core instructions.\n
- Capability inventory: Sub-agents spawned by the workflow have access to file system tools including read, grep, find, and ls.\n
- Sanitization: No sanitization, escaping, or validation is performed on the π.task input or retrieved file data before it is interpolated into prompts.\n- [COMMAND_EXECUTION]: The workflow programmatically executes file system tools (ls, find, grep, read) across sets of items discovered at runtime. While these tools are used for the skill's primary purpose of auditing and parallel research, the lack of input sanitization and prompt boundaries creates a risk that malicious data in the environment could influence the parameters or targets of these commands.
Audit Metadata