The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it ingests untrusted data from the sandbox environment (file names) and interpolates them into its output without sanitization or boundary markers.
Ingestion points: Directory contents retrieved from ctx.sandboxDir in handler.js.
Boundary markers: None present to delineate external file names from the summary template.
Capability inventory: The skill utilizes read_file and list_directory as defined in the allowed-tools frontmatter.
Sanitization: No escaping or validation is performed on the file names before they are joined into the result string.

summarise