The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from files such as RESOURCES.md, MISSION.md, and NOTES.md to guide its behavior. This creates a surface for indirect prompt injection if those files contain instructions from external or untrusted sources.
Ingestion points: MISSION.md, RESOURCES.md, NOTES.md, and learning records in SKILL.md.
Boundary markers: Not explicitly defined in the provided instructions.
Capability inventory: File system writes to ./lessons/ and ./assets/, and CLI command execution to open files.
Sanitization: No specific sanitization or validation logic is mentioned for the ingested data.
[COMMAND_EXECUTION]: The skill instructions suggest that the agent should open generated lesson files using CLI commands. While intended for user convenience, this involves the agent executing shell commands based on internal logic.
[DYNAMIC_EXECUTION]: The agent is tasked with generating HTML lessons and interactive components like simulators and quiz widgets in the ./assets/ directory. This involves the runtime generation and storage of executable web content (HTML/JavaScript) based on its training and the workspace context.

zj-teach