The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructions require the agent to ingest and analyze data from untrusted external sources, specifically call transcripts and email threads from sales deals. This creates a surface for indirect prompt injection where a buyer or third party could include malicious instructions within their communications to influence the agent's synthesis or findings. Ingestion points: Call transcripts, email threads, and proposal versions as described in Lane 1 of SKILL.md. Boundary markers: The skill does not provide specific instructions to use delimiters or safety tags when the agent processes these external records. Capability inventory: The agent is tasked with mining transcripts, extracting verbatim quotes, reconstructing decision timelines, and coding deals into a taxonomy. Sanitization: No sanitization, filtering, or validation steps are defined for the ingested deal records.

win-loss-program