The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes several Python scripts (e.g., batch_convert_templates.py, convert_to_csv.py) that use subprocess.call to execute local helper scripts. These calls are secure as they target internal scripts using the current Python interpreter (sys.executable) and do not enable shell execution (shell=False), preventing shell command injection from user-provided file paths or arguments.
[DATA_EXFILTRATION]: Files in JSON, CSV, Word, Excel, and XMind formats are parsed and transformed locally to support the QA workflow. No network activity, hardcoded secrets, or mechanisms for external data transmission were found in the transformation logic.
[PROMPT_INJECTION]: The skill processes user-supplied data, such as requirements and API documentation, through a library of prompts. This presents a surface for indirect prompt injection (Category 8). Evidence chain: Ingestion points include user requirements and design documents provided as context to prompts (e.g., requirements-analysis.md); boundary markers are absent; the capability inventory includes local file read/write and subprocess operations; sanitization of input data is not explicitly implemented. This surface is inherent to the skill's functional purpose and is managed by the agent's task-specific instructions.
[REMOTE_CODE_EXECUTION]: The skill relies entirely on the local execution of its provided scripts. No external dependencies are downloaded at runtime, and there is no pattern of executing scripts from remote or untrusted sources.

daily-testing-workflow