The Agent Skills Directory

[DATA_EXPOSURE]: The skill processes dataset schemas to identify sensitive fields. It includes explicit guardrails to avoid outputting real secrets, salts, or keys, and defaults to conservative actions (drop/redact) for high-risk fields like free-text.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data in the form of JSON schemas. While this represents a potential attack surface, the logic is confined to a local Python script that uses regular expressions for classification and does not interpolate data into executable commands or secondary LLM prompts in an unsafe manner.
Ingestion points: pseudonymization_field_mapper.py reads schema files from a user-defined path.
Boundary markers: None explicitly defined in the file format, but the logic uses structured JSON parsing.
Capability inventory: File system read/write via standard open() calls in pseudonymization_field_mapper.py.
Sanitization: Input is validated as JSON; field names are normalized via regex before classification.
[REMOTE_CODE_EXECUTION]: The provided Python script relies entirely on the Python standard library (argparse, json, re, dataclasses). It does not download external packages or execute remote code.

pseudonymization-field-mapper