The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to run a local script scripts/mobile_audit.py using the Bash tool. While the script itself is not provided in the input, the context of execution involves processing a user-provided <project_path>.
[REMOTE_CODE_EXECUTION]: The skill includes a highly suspicious instruction: "don't read, just run" regarding the mobile_audit.py script. This is an explicit attempt to bypass the agent's internal security inspection and code analysis capabilities before performing a command execution.
[PROMPT_INJECTION]: The markdown contains instructions that mandate specific "checkpoints" and "anti-memorization" files to be read. While primarily aimed at task performance, the rigid structure and "MANDATORY" labeling are designed to override the agent's default operational patterns.
[INDIRECT_PROMPT_INJECTION]: The skill creates a surface for indirect injection by reading data from a user-provided project directory.
Ingestion points: scripts/mobile_audit.py processes files from a directory provided by the user.
Boundary markers: None are present to distinguish between the script's instructions and the content of the files it audits.
Capability inventory: The skill has access to Bash, Read, Glob, and Grep tools, allowing it to modify the filesystem or execute commands based on input.
Sanitization: There is no mention of sanitization or validation for the data ingested from the <project_path>.

mobile-design