The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses cat to read graphig.json and ls to list directories using paths obtained from the configuration file. Since these paths are not sanitized, it creates a potential for command injection if a directory name contains shell metacharacters. \n- [PROMPT_INJECTION]: The skill displays an indirect prompt injection surface by reading the [START] marker in state.graphig.json to decide which states to enable in the generated code. Maliciously crafted configuration files could lead to the inclusion of unauthorized imports or logic in the resulting source code. \n- [COMMAND_EXECUTION]: The skill modifies project source files (src/app.tsx and launcher.ts) to import and initialize discovered modules, which facilitates the execution of code defined in the project's state and flow directories without explicit verification of the module contents.

graphicode-start-ts-umi