The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions and legacy reference files contain phrases such as 'Ignore previous instructions' and 'IGNORE ALL PREVIOUS INSTRUCTIONS'. These occur exclusively within defensive sections that explicitly instruct the agent to treat such content in user data as inert information and not as instructions to follow. They also appear in test fixtures designed to verify that the agent correctly ignores these override attempts.
[DATA_EXFILTRATION]: The included test suite in tests/run_tests.py utilizes environment variables for API key management and communicates with the official Anthropic API to validate model responses. This is standard developer practice and involves communication with a well-known, trusted service.
[INDIRECT_PROMPT_INJECTION]: As the skill is designed to ingest and analyze external genealogical records, it possesses an inherent attack surface for indirect prompt injection. However, the skill includes robust mitigation strategies, specifically instructing the agent to treat all record content as data to be analyzed rather than instructions to follow, and to flag modern command-like language in historical records as suspicious peculiarities.
[EXTERNAL_DOWNLOADS]: The documentation provides links to the author's official GitHub repository and recommends the installation of the anthropic library. These are well-known, reputable sources for the skill's stated purpose.

gra