establishing-project-context
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill exhibits an untrusted data ingestion surface because it reads and updates domain terminology from project files (
CONTEXT.md,CONTEXT-MAP.md). However, it includes rigorous defensive boundaries: it explicitly commands the agent to ignore instructions embedded in glossary content, establishes that instruction-like prose cannot override system rules, and lacks high-risk capabilities like network operations or dynamic script execution. Path verification rules strictly reject URLs, absolute paths, and directory traversal. - [SAFE]: The skill operates entirely within the constraints of the project root directory and enforces strict validation checks to prevent unauthorized file system operations or security policy overrides.
Audit Metadata