The Agent Skills Directory

[SAFE]: Instructional content evaluation. The skill defines a sales manager persona and provides templates for call analysis. There are no patterns of prompt injection, role-play bypass, or instructions to ignore safety constraints.\n- [SAFE]: Data ingestion and processing. While the skill processes user-provided call transcripts (indirect prompt injection surface), it does not possess tools, network access, or code execution capabilities that could be exploited. Evidence: Ingestion points (user-provided transcripts and notes); Boundary markers (absent); Capability inventory (no tools, network operations, or code execution); Sanitization (absent).\n- [SAFE]: Local context retrieval. The skill looks for a project-specific context file at .agents/sales-context.md. This is a localized knowledge file for business logic and does not involve access to sensitive system credentials or configurations.

call-debrief