The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to use the exec tool to run commands via the deeptutor binary.
[REMOTE_CODE_EXECUTION]: The skill has a dependency on an external binary deeptutor as specified in the requires.bins metadata. This is a vendor-owned resource for the authoring organization.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection (specifically command injection) by interpolating user input into a shell command template.
Ingestion points: Untrusted user input enters the system through the <problem description> placeholder in the deeptutor command.
Boundary markers: The input is wrapped in double quotes, which does not prevent command substitution or shell expansion in most shell environments.
Capability inventory: The skill leverages the exec tool which allows for shell-level execution on the host system.
Sanitization: No input validation or shell-escaping instructions are provided to mitigate the risk of command injection from the problem description.

deep-solve