The Agent Skills Directory

[SAFE]: The skill is built on an 'assume-malicious' posture toward its own data sources, mandating that the agent treats all records as potentially inaccurate until verified against multiple independent sources.
[SAFE]: Identity Resolution is implemented as a mandatory 'hard gate,' preventing any substantive research or reporting until a unique Identity Fingerprint is established to avoid namesake confusion.
[SAFE]: Strong privacy boundaries are enforced, explicitly prohibiting the collection of personal contact information (email, phone, address), the inference of protected traits (religion, health, politics), and the speculation of private motives for career transitions.
[SAFE]: The skill implements a robust 'Claim Confidence' vocabulary (Confirmed, Probable, Self-reported, Inferred, Unverified) and requires these classifications to be preserved across all output layers, including summaries and saved reports.
[SAFE]: High-risk agent behaviors are strictly controlled: parallel subagents require explicit secondary user permission, and the 'Saved Reports' feature is opt-in, project-local, and protected by overwrite confirmation prompts.
[SAFE]: The provided Python validation script (scripts/validate.py) is a benign tool designed for local static analysis of the skill's integrity, using only standard libraries and performing no network operations or external code execution.
[SAFE]: The skill explicitly refuses prohibited use cases such as formal background checks, suitability scoring, and lead generation, ensuring it remains within the scope of professional background research.

verified-person-research