The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the transfer of facts, commands, and constraints between agent sessions via the 'AGENTS.md' file, creating a persistent state that influences future model behavior. \n- Ingestion points: Agents are expected to read 'AGENTS.md' at the start of new tasks to gather context and instructions from previous runs. \n- Boundary markers: The skill does not provide any instructions for wrapping the 'standing lessons' in delimiters or adding security warnings to prevent agents from interpreting the text as direct command overrides. \n- Capability inventory: The skill utilizes file-writing capabilities to create or append to 'AGENTS.md' files within the repository. \n- Sanitization: There is no requirement or logic provided for the agent to sanitize the content of the lessons to remove potentially harmful instructions or prompt injection payloads before saving them.

compound