The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external Newick and Nexus files which represent an attack surface for indirect prompt injection. However, the skill treats these as structured data and provides explicit warnings in its documentation (references/api_reference.md and references/workflows.md) against using untrusted input with features like TreePattern that allow Python expressions.- Ingestion points: Newick/Nexus file reading via Tree(handle, parser=...) in scripts/tree_operations.py and scripts/quick_visualize.py.- Boundary markers: Not applicable for Newick format, but the parser enforces a specific grammar.- Capability inventory: File writing (scripts/tree_operations.py), network server initiation for visualization (scripts/quick_visualize.py), and Python code execution.- Sanitization: The skill contains clear instructions to avoid constructing expression-bearing patterns from untrusted sources.- [DYNAMIC_EXECUTION]: ETE 4 supports TreePattern search with embedded Python conditions. The skill's documentation explicitly identifies this as a potential risk and mandates that such patterns only be used with static, trusted code.- [COMMAND_EXECUTION]: The skill uses uv run to execute bundled Python scripts. These scripts are part of the skill and perform intended tree processing tasks with proper argument validation.- [DATA_EXFILTRATION]: The visualization script (scripts/quick_visualize.py) starts a local web server. It includes a validate_bind_address function that restricts the server to the loopback interface (127.0.0.1) by default, preventing unauthorized remote access unless the user provides explicit consent via --allow-remote-bind.

etetoolkit