agentic-coding-assistant
Warn
Audited by Gen Agent Trust Hub on Sep 5, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The TerminalIntegrationManager class and associated usage examples explicitly provide the capability to execute arbitrary shell commands, such as 'npm install', 'docker build', and other development workflows. This gives the agent high-level access to the host terminal environment.
- [INDIRECT_PROMPT_INJECTION]: The skill autonomously ingests codebase context and user instructions to plan and execute tasks, creating an attack surface where malicious code in the project files could influence the agent's behavior. 1. Ingestion points: 'task_description' in TaskPlanner and file contents in CheckpointRecoverySystem. 2. Boundary markers: No delimiters or ignore-instructions warnings are implemented. 3. Capability inventory: Terminal command execution, file read/write, and multi-file coordination. 4. Sanitization: No explicit sanitization or validation of codebase context before processing.
- [DYNAMIC_EXECUTION]: The Agentic Execution Engine dynamically generates multi-step execution plans and terminal workflows at runtime based on analyzed requirements and codebase context.
Audit Metadata