The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection because it processes untrusted data from an external source (AgentMail inbox) and presents it to the agent context.
Ingestion points: The scripts/scan_newsletters.py script retrieves messages from the AgentMail API using the agentmail client.
Boundary markers: The script does not utilize delimiters or specific instructions to the agent to disregard embedded commands within the extracted email text or snippets.
Capability inventory: The skill includes the send-email-via-agentmail capability, which could be targeted by a malicious email to perform unauthorized mailing actions if the agent follows instructions found in the data.
Sanitization: The script performs basic HTML stripping for formatting but lacks sanitization or filtering to detect or neutralize natural language prompt injection attempts.

newsletter-monitor