The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources and local user code, creating a surface for potential injection attacks.
Ingestion points: The skill fetches external content using WebFetch to vet primary sources (defined in SKILL.md) and reads user source code for "grounding" examples (defined in modes/lesson.md).
Boundary markers: The instructions do not specify the use of XML tags or other delimiters to isolate external content from the agent's instructions.
Capability inventory: The skill utilizes Write to create HTML lessons and exercises, Bash for repository management and exercise execution, and WebFetch for network access.
Sanitization: There is no explicit requirement for the agent to sanitize or escape data fetched from external URLs before interpolating it into lessons or exercises.
[COMMAND_EXECUTION]: The skill executes system commands to interact with the learning repository and launch generated content.
Evidence: The skill uses git init to initialize the learning repository at the specified path (~/learning).
Evidence: In SKILL.md, the "Opening the lesson" section instructs the agent to use platform-specific openers (open on macOS, xdg-open on Linux, start on Windows) to launch generated HTML files.
Evidence: In modes/lesson.md, the skill generates runnable code files in an exercises/ folder and prints the command required for the user to execute them.

tutorkit