maintain-verification-skill
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data by reading source code feature files and indexing project-local skill directories. * Ingestion points: Reads feature files from source code and sibling files in the feature map (SKILL.md). * Boundary markers: The instructions lack explicit delimiters or warnings to disregard potential injection attempts within the project files. * Capability inventory: Includes file reading, subagent spawning, and script execution. * Sanitization: No sanitization is performed on content read from the project before processing.
- [DYNAMIC_EXECUTION]: The skill executes harness scripts and launch models defined within target directories discovered via glob patterns at runtime. * Evidence: The 'Live pass' step instructs following the verification skill's own launch model, and 'Triage' involves ensuring scripts are executable. This poses a risk if a malicious target skill is audited.
- [COMMAND_EXECUTION]: The skill triggers the execution of external scripts and command-line tools as part of its feature verification and harness upkeep logic.
Audit Metadata