The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes data from the workspace and incorporates it into output formats like HTML and Markdown.
Ingestion points: Workspace files, line numbers, and API definitions are read from the current project to serve as evidence in the explanations.
Boundary markers: The instructions do not define markers or specific delimiters to separate ingested workspace content from the agent's own instructional logic.
Capability inventory: The skill can create a workspace directory and write files (HTML, Markdown, text). It is also instructed to open the generated HTML files in a web browser.
Sanitization: There are no instructions to escape or sanitize the workspace data before it is included in the output, which could allow malicious content in the workspace to inject scripts or instructions into the generated files.
[DYNAMIC_EXECUTION]: The skill dynamically generates an inline JavaScript script for a 'step-by-step player' when explaining time-ordered processes. Since this code is created at runtime and incorporates content derived from the subject being explained (which may involve untrusted workspace data), it poses a risk of script injection into a local file that the user is encouraged to open.

explain-clearly