Agent-first Python package for safe, scriptable image edits with GIMP.
- JSON-first CLI for LLM/tooling use
- Bridge-first architecture
- GIMP 3 batch execution backend
- Validation and deterministic outputs
pip install harnessgg-gimpharness-gimp bridge start
harness-gimp doctor
harness-gimp inspect input.jpg
harness-gimp resize input.jpg --width 1024 --height 768 --output resized.png
harness-gimp crop resized.png --x 0 --y 0 --width 800 --height 600 --output cropped.png
harness-gimp crop-center resized.png --width 800 --height 800 --output centered.png
harness-gimp export input.jpg output.webp
harness-gimp bridge statusAll commands print one JSON object to stdout.
Note: layer-edit commands are most reliable on .xcf working files.
Tip: prefer harness-gimp or harnessgg-gimp CLI entrypoints over python -m harness_gimp in mixed environments.
Tip: bridge start --port ... persists the bridge URL for later commands; override anytime with HARNESS_GIMP_BRIDGE_URL. Set HARNESS_GIMP_STATE_DIR to customize where bridge state files are stored.
- Human commands:
docs/human/commands.md - LLM quickstart:
docs/llm/quickstart.md - LLM command spec:
docs/llm/command-spec.md - LLM bridge protocol:
docs/llm/bridge-protocol.md - LLM response schema:
docs/llm/response-schema.json - LLM error codes:
docs/llm/error-codes.md