agent-context-management
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from tool outputs and external context artifacts, presenting a surface for indirect prompt injection. However, it includes explicit instructions for security mitigations:
- Ingestion points: Processes data from
read_file,search_code, anda2a-artifact.jsonsnapshots. - Boundary markers: Utilizes 'Smart Zones' (placing critical instructions at the top and bottom of the context window) and the 'Isolate' phase of its lifecycle to prevent poisoning.
- Capability inventory: The skill uses tools such as
execute_command,write_file, andedit_fileto manage codebase state. - Sanitization: Instructions explicitly mandate recording context provenance and sanitizing raw external data before it enters the primary reasoning buffer.
Audit Metadata