The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to analyze untrusted text for hidden signals, which inherently involves a surface for indirect prompt injection. The skill includes explicit instructions for the agent to treat detected Unicode tag characters or bidi overrides as high-severity injection risks.\n
Ingestion points: scripts/detect_text_watermark.py reads file paths or literal strings provided via command-line arguments.\n
Boundary markers: The skill does not wrap processed content in specific security delimiters.\n
Capability inventory: The skill performs local file reading and intensive Unicode analysis.\n
Sanitization: The skill identifies but does not automatically strip malicious content; it relies on reporting results to the agent.\n- [DYNAMIC_EXECUTION]: The vendored shared library scripts/provenance_core.py includes a run_tool function that uses subprocess.Popen with shell=False. This function is not utilized by the current skill's text analysis workflow and follows secure implementation patterns.

detect-text-watermark