deep-agents-core

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [Filesystem Capabilities]: The skill describes the use of FilesystemMiddleware and FilesystemBackend which provide agents with tools such as read_file, write_file, and edit_file. These capabilities allow agents to manage project context and persistent data within a specified directory.
  • [Dynamic Skill Loading]: The documentation outlines a SkillsMiddleware that dynamically loads SKILL.md files from a local directory or a store backend. This allows agents to adopt specialized capabilities on-demand based on the task at hand.
  • [Sub-Agent Delegation]: The framework includes a SubAgentMiddleware that enables agents to spawn and manage specialized sub-agents. This delegation pattern is used to break down complex tasks into smaller, manageable units.
  • [Indirect Prompt Injection Surface]: Because the agents created with this framework are designed to ingest and process external files and user input, they possess an inherent surface for indirect prompt injection. Developers are encouraged to consider boundary markers and sanitization when implementing these agents in production environments.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 11:51 AM
Security Audit — agent-trust-hub — deep-agents-core