The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its reliance on data from external tool outputs and user-provided inputs without explicit safety boundaries. * Ingestion points: The agent processes data from the helm-asc CLI output (such as uploadsInbox paths, perGroup counts, and items metadata) as well as design tokens and filenames provided by the user. * Boundary markers: The instructions lack specific delimiters or instructions for the agent to distinguish between data and potential embedded commands within these inputs. * Capability inventory: The skill involves the use of several shell commands, including helm-asc, find, mkdir, cp, and rm. * Sanitization: There are no specified procedures for validating or escaping the data obtained from external tool outputs or filenames before they are used in subsequent operations.
[COMMAND_EXECUTION]: The skill requires the execution of multiple shell commands to perform its tasks. It utilizes a specialized CLI tool (helm-asc) and standard filesystem utilities (find, mkdir, cp, rm). While these are functional requirements, they represent powerful capabilities that could be leveraged if the agent receives malicious instructions through indirect channels.
[DATA_EXFILTRATION]: The skill accesses a specific application data path at ~/Library/Group Containers/group.com.modumhq.Helm. Accessing specific application directories for data staging and upload involves reading potentially sensitive application data, which could be exposed if the agent context is manipulated.

asc-screenshots