The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data in the form of conversation threads (Slack, email, or meeting transcripts).
Ingestion points: The SKILL.md file defines 'The thread' as a required input, which is where untrusted external content enters the agent's context.
Boundary markers: The instructions require the agent to use verbatim quotes for every attribution, which serves as a natural boundary between the source data and the agent's analysis.
Capability inventory: The skill is restricted to text analysis and markdown report generation. It does not utilize any file-system writes, network operations, or subprocess execution tools.
Sanitization: The 'Forensic Protocol' and 'Anti-Patterns' sections emphasize accuracy and the surfacing of ambiguity, which reduces the likelihood of the agent being manipulated by instructions embedded within the source threads, though it does not programmatically sanitize the input.

decision-forensics