structured-spec
Pass
Audited by Gen Agent Trust Hub on Oct 18, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines a system for parsing and validating specification documents that include natural language requirements and executable code contracts. This structure creates an attack surface where malicious instructions could be embedded in requirement fields (such as rationale or acceptance criteria) to influence agent behavior during document review, validation, or the multi-agent slicing orchestration process.\n
- Ingestion points: Markdown specification files containing HTML-comment annotations (requirement, contract, decision, slo, etc.) as specified in specification.md.\n
- Boundary markers: The standard utilizes specific HTML comment tags (e.g., ) to separate structured metadata from general narrative prose.\n
- Capability inventory: The agent is instructed to write, review, and validate documents, and the standard supports a workflow for partitioning tasks to sub-agents based on the parsed manifest.\n
- Sanitization: The skill instructions do not explicitly direct the agent to sanitize or filter the natural language content within annotations for potential prompt injection or adversarial instructions before processing.
Audit Metadata