context-management-context-save
Pass
Audited by Gen Agent Trust Hub on Oct 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data by analyzing the entire project directory ($PROJECT_ROOT) to extract architectural patterns, decision rationales, and metadata. Maliciously crafted content within project files (e.g., in documentation, comments, or configuration files) could trigger unintended agent behavior during context serialization.
- Ingestion points: File content and metadata extracted from the directory specified in $PROJECT_ROOT.
- Boundary markers: The instructions do not define clear delimiters or use 'ignore' directives to prevent the agent from obeying instructions embedded within the analyzed project files.
- Capability inventory: The skill performs extensive filesystem reads, semantic analysis, and external storage operations (Vector Database integration with Pinecone, Weaviate, and Qdrant).
- Sanitization: While the documentation mentions that sensitive information should be excluded, no technical sanitization, filtering, or validation mechanism is implemented within the skill logic to scrub external content.
Audit Metadata