The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill facilitates the activation of GIMP's Script-Fu server, creating an unauthenticated TCP listener on 127.0.0.1:10008. This server accepts and executes arbitrary Python or Scheme code, allowing any process on the local machine to execute commands with the user's privileges within the GIMP environment.
[DYNAMIC_EXECUTION]: The file scripts/gimp_client.py dynamically generates a Python bootstrap script at runtime. It uses the exec(compile(...)) pattern to execute agent-generated scripts within GIMP's internal Python interpreter, bypassing static analysis of the executed logic.
[COMMAND_EXECUTION]: The skill relies on shell scripts (scripts/gimp-send.sh and scripts/gimp-start.sh) to launch GIMP binaries and manage the transmission of code to the running instance.
[INDIRECT_PROMPT_INJECTION]: The skill translates natural language requests into executable Python scripts. This creates a vulnerability surface where malicious instructions embedded in processed data (such as attacker-controlled filenames, image metadata, or folder paths) could be interpolated into the generated scripts.
Ingestion points: User-supplied filenames, paths, and text content interpolated into GIMP scripts (seen in scripts/example-poster.py and references/recipes.md).
Boundary markers: The instructions do not define strict boundary markers to distinguish between trusted code templates and untrusted user data.
Capability inventory: The skill has full access to the GIMP API (file export, layer manipulation) and standard Python libraries (file system access via os and glob).
Sanitization: There is no evidence of escaping or sanitization for external content before it is interpolated into the Python code delivered to GIMP.

gimp