ai-cost-optimizer
Pass
Audited by Gen Agent Trust Hub on Oct 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill provides implementation patterns for ingesting large external datasets, such as full source code repositories, which serves as a potential vector for indirect prompt injection if the content is not properly sanitized.
- Ingestion points: The implementation patterns in
SKILL.mdandreferences/2-advanced-caching.mdinvolve passingfullRepoDataand long documents into the model's context. - Boundary markers: The skill proactively recommends using XML tags (e.g.,
<system_instruction>,<role>,<constraints>) to delimit instructions and mitigate instruction drift. - Capability inventory: The skill provides conceptual code snippets for model generation and cache management but does not include active scripts or executable tools with file-system or network access.
- Sanitization: The instructions emphasize 'Token Engineering' and the use of explicit 'Do Not' lists to maintain control over model behavior when processing large contexts.
Audit Metadata