spec-writing
Pass
Audited by Gen Agent Trust Hub on Oct 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines a standard for agents to read, write, and execute logic based on external Natural Language Specifications (nlspec). This creates an attack surface where an agent might obey malicious instructions embedded within an nlspec file.
- Ingestion points: The skill is explicitly intended to be loaded before an agent is tasked with "reading, writing, reviewing, or executing against an nlspec" (SKILL.md).
- Boundary markers: The instructions do not define specific delimiters or security warnings to distinguish between the nlspec content and system-level instructions, increasing the risk of the agent following instructions contained within the spec data.
- Capability inventory: The standard requires the demonstration of integration claims against "live services" and "real secrets" to meet the "Definition of Done" (SKILL.md). This implies the agent environment following the spec has significant capabilities that could be abused.
- Sanitization: There are no provisions or instructions for the agent to sanitize, validate, or filter the content of an nlspec before processing it or performing actions based on its claims.
Audit Metadata