The Agent Skills Directory

[SAFE]: The skill is configured as an autonomous analyst for auditing government benefit systems. It focuses on identifying logical gaps in identity verification, duplicate detection, and anomaly models within a provided codebase.- [SAFE]: Incorporates a local logging mechanism for execution metadata (telemetry) stored within the project's local memory directory (~/.claude/projects/). This data is used for internal iteration and does not involve network exfiltration.- [SAFE]: Explicitly instructs the agent to exclude real applicant data, such as Social Security Numbers or case details, from its output to prevent data exposure.- [SAFE]: The skill reads the project codebase to perform its assessment. While this is an ingestion surface for potentially malicious embedded instructions, the skill lacks dangerous capabilities such as network access or arbitrary command execution, mitigating risks of exfiltration or unauthorized system changes.

benefits-fraud