The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the local file system and CLI tool outputs to determine upload actions, creating a potential surface for indirect instructions.
Ingestion points: The agent reads repository folder names, filenames (e.g., NN_<filename>.<ext>), and display type tokens which are used to build the staging tree. It also parses output from helm-asc version ... screenshots and helm-asc ... upload --dry-run.
Boundary markers: The skill uses structured CLI flags (e.g., --agent, --dry-run, --path) and specific directory structures to manage data, but it does not specify explicit delimiters to isolate external metadata from instructions.
Capability inventory: The skill possesses the ability to create directories (mkdir), copy files (cp), list files (find), and perform network-based uploads via the helm-asc command.
Sanitization: There is no explicit sanitization described for the filenames or the content returned by the helm-asc tool before it is processed by the agent.

asc-screenshots