context-management-context-save
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: No security issues detected. The skill functions as a reference for context engineering and includes instructions to avoid capturing sensitive data and secrets.
- [INDIRECT_PROMPT_INJECTION]: The skill describes processes for ingesting and summarizing project data, which constitutes a potential attack surface for indirect prompt injection from files within a repository.
- Ingestion points: Data is extracted from the directory defined by the
$PROJECT_ROOTparameter (SKILL.md). - Boundary markers: The instructions explicitly advise users to "define allowed roots/data" and "Exclude secrets and unrelated personal content" (SKILL.md).
- Capability inventory: The skill outlines logic for semantic extraction, JSON serialization, and storage in external databases like Pinecone or Weaviate.
- Sanitization: It recommends the explicit exclusion of sensitive information and suggests that captured context be treated as historical data rather than authority.
Audit Metadata