The Agent Skills Directory

[SAFE]: No malicious patterns or behaviors were identified. The skill operates entirely through natural language instructions and logical templates.
[NO_CODE]: The skill does not contain any executable scripts, binary files, or installation dependencies. All identification logic is described as text-based instructions for the agent.
[PROMPT_INJECTION]: The skill processes untrusted input data from reimbursement records and applicant history, representing a theoretical surface for indirect prompt injection. However, since the skill possesses no sensitive capabilities (e.g., file-system writes or network access), this surface does not pose a functional risk.
Ingestion points: Data enters the agent context via the argument-hint parameters (SKILL.md).
Boundary markers: No specific delimiters or 'ignore embedded instructions' directives are employed.
Capability inventory: The skill is limited to logic analysis and structured report generation.
Sanitization: No data validation or escaping routines are present.
[DATA_EXFILTRATION]: The skill references a configuration file at ../../CLAUDE.md. While accessing files in a parent directory is a best-practice concern regarding directory traversal, the behavior is used solely for loading local configuration and does not involve external data transmission.

dispute-identification