The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process content from external sources (documented in RESOURCES.md) and user-provided inputs (MISSION.md, NOTES.md) to create lessons. If an external resource contains malicious instructions, they could potentially influence the agent's output or be injected into the generated HTML files. Ingestion points: Data is read from RESOURCES.md, MISSION.md, NOTES.md, and files in learning-records/. Boundary markers: The instructions do not specify the use of delimiters or warnings to ignore instructions embedded in the external content. Capability inventory: The agent has the capability to write files (.html, .md, .css, .js) and execute shell commands to open files. Sanitization: There are no explicit instructions to sanitize or validate external content before it is interpolated into lessons or assets.
[DYNAMIC_EXECUTION]: The skill involves the generation of reusable assets such as stylesheets and interactive widgets (JavaScript) in the ./assets/ directory. This constitutes runtime script generation which, while intended for interactive learning, could be a vector for cross-site scripting (XSS) if influenced by malicious input.
[COMMAND_EXECUTION]: The instructions suggest that the agent should use CLI commands to open generated lesson files for the user. This involves shell interaction which, although limited to opening local files, is a capability that should be monitored for potential path traversal or command injection if file names are derived from untrusted input.

teach