ulw-loop
Pass
Audited by Gen Agent Trust Hub on Sep 27, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill orchestrates a variety of shell-based utilities for quality assurance and repository management, including git for version control operations, curl for HTTP endpoint validation, and tmux for terminal session management. These tools are leveraged to provide observable evidence for goal completion.
- [DYNAMIC_EXECUTION]: The skill uses a dynamic import mechanism to load the agentToolkit SDK at runtime. The import path is constructed using the OMO_AGENT_TOOLKIT_SDK_ROOT environment variable, which is a standard pattern for integrating with the host platform's extension services.
- [INDIRECT_PROMPT_INJECTION]: The skill manages a workflow where untrusted data from multiple sources is ingested into a high-capability execution environment, creating a potential surface for indirect prompt injection.
- Ingestion points: User-provided 'brief' text processed by the createGoals method and output results from subagents dispatched via the task tool.
- Boundary markers: The instructions specify the use of TASK: prefixes for subagent prompts but do not define explicit delimiters or ignore instructions for sanitizing user inputs within SDK calls.
- Capability inventory: The skill has access to sensitive capabilities including file system modification (git worktrees), network communication (curl, playwright), and process management (kill, tmux).
- Sanitization: The provided instructions do not include specific sanitization or validation logic for data processed during the goal execution loop.
Audit Metadata