agent-workflow-patterns
Pass
Audited by Gen Agent Trust Hub on Oct 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill provides high-level design patterns for composing LLM systems (e.g., prompt chaining, routing, and parallelization). It contains no executable scripts, external dependencies, network operations, or sensitive file access.\n- [INDIRECT_PROMPT_INJECTION]: The workflow patterns described in the skill are intended to process untrusted user input, which represents a potential attack surface. The skill proactively addresses this by including specific security instructions, such as validating routing outputs against allowlists and enforcing schema validation on worker results.\n
- Ingestion points: User-supplied input processed by routing, chaining, and worker steps in
SKILL.md.\n - Boundary markers: Recommends validating classifier outputs and using non-droppable intent/constraint fields across workflow steps.\n
- Capability inventory: None; the skill defines architectural logic and does not provide direct system, tool, or network capabilities.\n
- Sanitization: Advises on schema validation and red-flag-discarding of malformed or suspicious worker outputs.
Audit Metadata