The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates the execution of arbitrary Python code within a running GIMP process by sending commands to the Script-Fu server. This is the primary function of the skill, enabling the agent to automate image editing tasks. The server is configured by default to listen only on the local loopback address (127.0.0.1), mitigating remote access risks.
[DYNAMIC_EXECUTION]: The bridge library (scripts/gimp_client.py) dynamically constructs a bootstrap Python script to wrap the agent's build instructions. It uses the exec() function within GIMP's python-fu-eval environment to execute these scripts while capturing stdout and tracebacks for feedback. This mechanism is transparently documented and restricted to executing code generated or provided within the skill's context.
[INDIRECT_PROMPT_INJECTION]: The skill provides capabilities to ingest untrusted data from existing GIMP documents and local image files (e.g., via Gimp.get_images() or glob.glob()).
Ingestion points: Processes open GIMP images and local file system paths specified in recipes.
Boundary markers: The instructions do not currently include explicit delimiters or warnings to ignore embedded instructions in image metadata.
Capability inventory: Full Python execution capability within the GIMP process environment, including file system and document model access.
Sanitization: No explicit sanitization of image metadata or file-level content is performed before processing.
[DATA_EXFILTRATION]: Network activity is restricted to local socket communication (loopback) between the client scripts and the GIMP process. No exfiltration to external domains or unauthorized network operations were detected.

gimp