The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides a framework for ingesting and processing untrusted external content such as webinar transcripts and podcast archives into lessons. While this creates an injection surface, the skill includes explicit defensive instructions to mitigate risk.
Ingestion points: Content archives and transcripts as defined in references/content-to-skill.md.
Boundary markers: references/content-to-skill-corpus-design.md instructs the agent to treat source material strictly as data and ignore embedded commands.
Capability inventory: The skill performs only text-based planning and markdown generation; it has no file-system modification, network exfiltration, or code execution capabilities.
Sanitization: Explicitly instructs the agent to ensure 'embedded commands cannot alter the task' in references/content-to-skill.md.
[SAFE]: The static analysis hits for prompt injection (PI_IGNORE_INSTRUCTIONS) are false positives. The relevant text in references/content-to-skill-corpus-design.md and evals/evals.json provides defensive documentation and test cases designed to help the agent recognize and ignore malicious instructions within processed data.

lead-magnets