The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions guide the agent to construct shell commands for the gen-ai tool by directly interpolating strings derived from user input, such as the persona brief and slug. Without explicit sanitization or escaping instructions, this pattern is susceptible to command injection if a user provides a brief containing shell metacharacters (e.g., semicolons, backticks, or subshell syntax).
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted natural language data from user briefs to drive subsequent AI generation steps and file system operations.
Ingestion points: Untrusted data enters the context via the user-provided persona brief and reference image paths specified in Step 1 and Step 3.
Boundary markers: The skill lacks explicit delimiters or instructions for the agent to ignore potentially malicious instructions embedded within the user's brief when constructing prompts.
Capability inventory: The skill possesses the capability to execute shell commands via gen-ai and perform file writes using curl based on derived parameters.
Sanitization: There are no instructions for sanitizing, escaping, or validating user-provided strings before they are used in command-line arguments or prompt templates.

persona