The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data such as source code and conversation history to generate its explanations. This represents a potential surface for indirect prompt injection if the processed data contains malicious instructions, though this ingestion is fundamental to the skill's primary educational purpose.
Ingestion points: The skill reads local source code files and previous conversation history (SKILL.md, step 2).
Boundary markers: No specific delimiters or "ignore previous instructions" guards are implemented for the agent when it reads external code.
Capability inventory: The skill can read code, invoke internal sub-skills (how, why, unslop), and utilize an image-generation tool.
Sanitization: There are no instructions for sanitizing or escaping the external content before it is processed by the LLM.
[SAFE]: The skill includes a reference to a public repository on GitHub (github.com/cursor/plugins) for license attribution. As GitHub is a well-known service and the link is for documentation purposes, this reference is considered safe.
[SAFE]: The skill uses YAML frontmatter (disable-model-invocation: true) and OpenAI configuration (allow_implicit_invocation: false) to implement security restrictions that limit how and when the skill can be triggered, reducing the risk of unintended execution.

teach