The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's core workflow relies on ingesting content from external sources to build lessons. This creates a vulnerability where malicious instructions could be embedded in the resources the agent is told to trust.
Ingestion points: The agent is instructed to find and read external resources, articles, books, and community forums (e.g., Reddit) to populate RESOURCES.md and draft lessons.
Boundary markers: There are no instructions for the agent to use delimiters or to disregard embedded instructions within the external content it retrieves.
Capability inventory: The agent has the capability to write multiple file types (.html, .md) to the workspace and execute CLI commands to open these files.
Sanitization: The skill lacks instructions for sanitizing or escaping content retrieved from external URLs before interpolating it into the generated HTML lessons or Markdown records.
[COMMAND_EXECUTION]: The instructions specifically encourage the use of shell commands to facilitate the user workflow.
Evidence: The SKILL.md file states: 'If possible, open the lesson file for the user by running a CLI command.' This confirms the agent is expected to use system-level commands to interact with the generated files.

light-teach