The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process information from various external and user-controlled sources, creating a pathway for malicious instructions to influence agent behavior. \n
Ingestion points: Untrusted content may enter through the MISSION.md goal statement, GLOSSARY.md definitions, and curated external links in RESOURCES.md. \n
Boundary markers: The instructions do not define any specific delimiters or constraints to prevent the agent from executing instructions that may be embedded within these materials. \n
Capability inventory: The agent has the authority to write files to the workspace and is encouraged to execute CLI commands, which increases the potential impact of an injection. \n
Sanitization: No logic is provided to ensure that input from external resources or user-provided files is sanitized or validated before being used by the agent to construct lessons or logic. \n- [COMMAND_EXECUTION]: The skill explicitly instructs the agent to 'run a CLI command' to open lesson files. This functionality introduces the capability for shell execution which, while intended for local file interaction, could be exploited if the agent is manipulated into executing unsanitized paths or malicious arguments.

teach