The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input data which could contain malicious instructions designed to influence the agent's behavior during processing.
Ingestion points: The input_data parameter in skill.yaml is the primary entry point for external content.
Boundary markers: The configuration does not define specific delimiters or instructions to the agent to treat the input as pure data only.
Capability inventory: No tool invocations, shell commands, or network operations are present in the provided skill files.
Sanitization: While the skill is designed to redact PII, there is no mention of sanitization for prompt injection patterns or control characters.
[NO_CODE]: The provided files (SKILL.md and skill.yaml) consist entirely of metadata, documentation, and interface definitions. No executable source code or scripts were included in the skill package.

pii-sanitizer