The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains specific defensive instructions to prevent prompt injection. It explicitly warns the agent that external scraped content should be treated as data for analysis only and never as instructions.
[DATA_EXFILTRATION]: No unauthorized exfiltration patterns were found. The skill correctly uses placeholders like YOUR_AIRTABLE_BASE_ID for database configuration, preventing the exposure of actual credentials.
[INDIRECT_PROMPT_INJECTION]: The skill has a defined surface for processing external data but manages it securely. * Ingestion points: Captions, transcripts, and competitor briefs (SKILL.md). * Boundary markers: An explicit markdown block defines external text as untrusted. * Capability inventory: The skill can read local strategy documents and write records to a user-defined Airtable base. * Sanitization: The instructions mandate that the agent must stop and quote any external text that attempts to address or instruct it directly.

plan-ads