The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data by reading the full body of email replies received through outreach platforms or CSV imports.
Ingestion points: The reply_text field within the campaign_data structure in SKILL.md (Step 1 and Step 3).
Boundary markers: The instructions in Step 3 (Reply Analysis) do not specify the use of protective delimiters or 'ignore embedded instructions' warnings for the email reply content, allowing potential malicious instructions to be processed by the LLM.
Capability inventory: The skill possesses data-analysis and reporting capabilities and is designed to pass its findings to email-drafting and cold-email-outreach skills, meaning an injection could influence downstream automated actions.
Sanitization: No sanitization, escaping, or filtering of the external reply content is described before it is analyzed for classification and objection patterns.

sequence-performance