The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and parses content from various external file formats (Excel, Word, CSV, XMind, Markdown) using custom scripts like scripts/parse_formats.py and scripts/xlsx_reader.py. This presents an attack surface where an attacker could embed malicious instructions in a test document to manipulate the agent's release recommendations or subsequent skill calls.
Ingestion points: scripts/parse_formats.py, scripts/convert_formats.py, and scripts/xlsx_reader.py (via read_workbook).
Boundary markers: Absent; the agent is not instructed to treat parsed content as untrusted data.
Capability inventory: The skill generates Go/No-Go decisions and orchestrates other AI agent skills.
Sanitization: The scripts perform no sanitization of the extracted text before it enters the model's context.- [DYNAMIC_EXECUTION]: Several utility scripts in the scripts/ directory (e.g., convert_to_csv.py, parse_excel.py, batch_convert_templates.py) use subprocess.call to dynamically invoke other Python scripts within the package. While these calls target local files using computed paths, they facilitate the execution of logic based on command-line arguments.- [COMMAND_EXECUTION]: The skill package includes numerous Python scripts that execute other local processes using subprocess.call. This is used to pipe data through various conversion stages.- [PRIVILEGE_ESCALATION]: The README.md documentation encourages Windows users to run a PowerShell installation script with the -ExecutionPolicy Bypass flag, which disables the security feature intended to prevent the execution of untrusted scripts.

release-testing-workflow