checking-gates
Pass
Audited by Gen Agent Trust Hub on Sep 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests metadata from tasks which can be influenced by external sources.
- Ingestion points: The
TaskGettool is used in SKILL.md to load task descriptions and parse thejson:metadatafence. - Boundary markers: The instructions do not specify any boundary markers or instructions to ignore malicious content within the metadata.
- Capability inventory: The skill has the capability to execute shell commands (via
verifyCommand), dispatch subagents, and update task statuses. - Sanitization: There is no evidence of sanitization or validation of the
verifyCommandstring before it is passed to the shell. - [DYNAMIC_EXECUTION]: The skill uses the
verifyCommandfield from task metadata to execute arbitrary shell commands at runtime. - Evidence: Step 3 explicitly states "Run the verifyCommand (or dispatch the subagent with subagentBrief). Capture exact output."
- [COMMAND_EXECUTION]: Arbitrary command execution is triggered based on the contents of task metadata, presenting a risk of unauthorized command execution if the task source (metadata) is manipulated by an external source.
Audit Metadata