The Agent Skills Directory

[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection by ingesting and processing untrusted user data to generate code.\n
Ingestion points: Workflow Step 1 in SKILL.md requires the agent to infer product context and audience requirements from user input.\n
Boundary markers: No specific boundary markers or instructions to treat the user-provided data as potentially untrusted are present in the workflow.\n
Capability inventory: The agent has the capability to write local HTML/JS files (SKILL.md, Step 3) and open them in a browser tool for interaction (SKILL.md, Step 4).\n
Sanitization: The instructions do not define requirements for sanitizing or escaping the user-provided context before it is interpolated into the generated playground code.\n- [COMMAND_EXECUTION]: The skill involves the generation of code and the execution of system tools as part of its primary function.\n
Code Execution: SKILL.md Workflow Step 4 instructs the agent to open and interact with the generated HTML file in a browser environment to verify its behavior.\n
Tool Usage: The templates/diff-review.md template suggests using the git show command to retrieve real commit data for use within the generated playground.

playground