The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/gen.mjs uses the jiti library to dynamically import and execute a file (../../../src/skill-cli.ts) located outside of the skill's package structure. This creates a dependency on the host environment's directory layout and could potentially result in the execution of unintended files if the environment is manipulated.
[DATA_EXFILTRATION]: The scripts/gen.mjs file passes the entire process.env object to the runImageCli function. This exposes all environment variables in the agent's context, including potentially sensitive system tokens or credentials unrelated to the skill, to the logic contained in the imported library.
[PROMPT_INJECTION]: The skill ingests natural language via the --prompt argument and produces external output (URLs), representing an indirect prompt injection surface.
Ingestion points: --prompt parameter in scripts/gen.mjs.
Boundary markers: None identified in the provided wrapper script.
Capability inventory: Command execution (Node.js) and network operations (image generation service).
Sanitization: No explicit sanitization or validation of prompt content is performed within the script wrapper.

coze-image-gen