The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the execution of a local Python script to perform its anonymization tasks. The agent is instructed to invoke this script through the system shell.\n
Evidence: The SKILL.md file contains explicit instructions for the agent to run the anonimizar.py script using the Python interpreter.\n
Evidence: The test_anonimizar.py script uses the subprocess.run function to execute the main tool during regression tests, following secure implementation parameters.\n- [PROMPT_INJECTION]: The skill ingests and processes untrusted files (CSV, XLSX, MD, DOCX), which represents an attack surface for indirect prompt injection if those files contain malicious instructions or specifically crafted token collisions.\n
Ingestion points: External data is ingested via document parsing libraries (pandas, openpyxl, python-docx) in the anonimizar.py script.\n
Boundary markers: The skill includes token collision detection logic (_PATRON_TOKEN_FUENTE) to identify if source files already contain strings mimicking its internal token format.\n
Capability inventory: The skill possesses capabilities for local filesystem read/write and script execution via the Python interpreter.\n
Sanitization: The tool focuses on sanitizing PII content but relies on the security of its dependency libraries for the safe parsing of document structures.\n- [EXTERNAL_DOWNLOADS]: The skill's setup instructions involve downloading pre-trained language models from the official spaCy registry.\n
Evidence: Installation steps in the documentation reference the use of python -m spacy download to fetch necessary language model data.

anonimizar