The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes non-recursive directory listings (ls -1) and path existence checks to identify canonical subtrees within the BSP and source directories. These operations are limited to one level of depth and are used solely for metadata discovery.
[DATA_EXFILTRATION]: The skill explicitly forbids network activity. It is designed to operate offline, and instructions specifically mandate that remote URLs found in the profile should be recorded verbatim without being fetched.
[PROMPT_INJECTION]: The skill processes external data including filenames from the local filesystem and descriptive strings from configuration files. While this creates an indirect prompt injection surface when the data is rendered into the output markdown, the risk is mitigated by the skill's restricted scope and the lack of high-privilege capabilities associated with the rendered output.

jetson-generate-kb