The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data which serves as an ingestion point for untrusted content. The generated profiles, particularly the top_values field for string columns, could contain adversarial instructions intended to influence the agent's behavior during subsequent tasks.
Ingestion points: DataFrames provided to profile_dataframe and other functions in scripts/profiling.py.
Boundary markers: Absent. The skill instructions do not provide delimiters or guidance for the agent to treat the profile output as untrusted data.
Capability inventory: The skill operates in an environment with high-privilege tools including Bash, Read, Write, and Edit as specified in SKILL.md.
Sanitization: Absent. The profiling logic extracts and displays raw string values from the dataset without filtering or escaping.

data-profiler