alterlab-skill-finder
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input to select appropriate specialized tools and to generate detailed task specifications for subagents. This represents a potential surface for indirect injection if downstream skills do not handle generated task specs securely.
- Ingestion points: User queries processed in the 'Route' and 'Orchestrate' modes defined in SKILL.md.
- Boundary markers: The instructions explicitly mandate giving subagents a 'complete task spec' with 'clear boundaries'.
- Capability inventory: The skill uses the 'Read Task' tool and has the capability to dispatch tasks to a library of 239 specialized academic skills or spawn subagents using host workflow tools.
- Sanitization: The skill enforces a human-in-the-loop safety gate, requiring mandatory clarifying questions and a plan confirmation step before any execution starts.
- [DYNAMIC_EXECUTION]: The skill dynamically assembles multi-agent workflows and generates task specifications at runtime based on user-provided goals. This behavior is a core functional requirement for the orchestrator-worker pattern but involves dynamic logic assembly that depends on interpreting external input.
Audit Metadata