The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill includes code examples (e.g., react_agent in SKILL.md and references/examples.md) that use Python's eval() function to process model-generated text. This pattern executes arbitrary strings as code, which is dangerous if the model output is compromised.
[COMMAND_EXECUTION]: The use of eval() in provided tool examples creates a vector for arbitrary code execution within the Python environment.
[INDIRECT_PROMPT_INJECTION]: The skill demonstrates processing external data which could contain malicious instructions.
Ingestion points: Functions such as extract_entities(lm, text) in SKILL.md and references/examples.md ingest external text strings.
Boundary markers: External data is delimited by simple headers like Text: {text} without robust isolation or instruction-bypass protection.
Capability inventory: The documented environment permits code execution through the eval() tool example.
Sanitization: No input sanitization or instruction filtering is demonstrated for the processed text before it is passed to the generation loop.

guidance