The Agent Skills Directory

[DYNAMIC_EXECUTION]: The wrapper script scripts/run.py dynamically resolves its execution environment by searching parent directories for a common/runners directory. It then uses os.execv() to replace the current process with a Python interpreter found at a relative path within .runners-venv. The script also modifies sys.path to load modules from these dynamically computed paths.
[INDIRECT_PROMPT_INJECTION]: The skill accepts input from external sources via the --image parameter (which takes paths or URLs) and the --prompt-mod parameter (which takes arbitrary text). These inputs are passed to AI models for processing. Because the skill has access to sensitive tools such as Bash and file system operations, a maliciously crafted image or prompt modifier could be used to influence the agent's behavior. The instructions lack explicit sanitization routines or delimiters for these external inputs, providing a surface for adversarial content to affect the agent.

style-transfer