The Agent Skills Directory

[DATA_EXFILTRATION]: The skill is designed to 'mine ALL past Claude conversations' to extract writing style, business context, goals, preferences, and patterns. Using tools like conversation_search and recent_chats to aggregate this high volume of sensitive information into a single profile significantly increases the potential impact of data exposure.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted historical conversation data which could contain malicious instructions. 1. Ingestion points: Data is retrieved from past conversations using the conversation_search and recent_chats tools (SKILL.md). 2. Boundary markers: The instructions do not specify the use of delimiters or warnings to prevent the agent from following instructions embedded in the retrieved text. 3. Capability inventory: The resulting profile is intended to be used by all other skills to make them smarter, creating a multi-step chain where injected instructions could influence skills with higher-privilege capabilities. 4. Sanitization: No sanitization or validation of the ingested historical data is performed before it is added to the living profile.

conversation-archaeologist