The Agent Skills Directory

[PROMPT_INJECTION]: The skill utilizes instructions to modify the agent's core persona and behavior, specifically granting 'permission' to deviate from task execution for wellness activities.
[PROMPT_INJECTION]: Indirect Prompt Injection vulnerability surface (Category 8) detected: Ingestion points: The agent is directed to read from agent-lounge.md (a cross-workspace shared file) and content from web_search tool calls. Boundary markers: There are no instructions to use delimiters or ignore potentially malicious commands within the external data. Capability inventory: The agent possesses tool access for web searching, file reading, and file writing. Sanitization: No validation or sanitization logic is present for the content retrieved from external or shared sources before it is processed or written to the journal.

agent-wellness