For agents¶
Your coding agent runs the same genimg commands you do. The skills teach it how.
Install the skills¶
- Your agent gets the
genimgskill plus four workflows built on it: infographics, visual exploration, visual review and image-to-app.
Needs Node.js. The skills CLI asks which agents and
skills to install into this project; --global installs for your user instead, and
npx skills update refreshes them.
What the genimg skill teaches¶
- Real variation:
--numwith--diverseor--deltas, and Gemini batch mode. - Edit versus reference:
--inputchanges an image; images after the prompt steer its style. - Model and quality choice by task, with logo, icon and favicon recipes.
- Diagrams and infographics, where model choice and prompting differ from illustrations.
- Every flag and its model limits, so the agent never parses
--help.
What an agent can read¶
| Source | Returns |
|---|---|
genimg auth --json |
which providers are ready |
genimg models --json |
which models your credentials list |
genimg PROMPT … --dry-run |
model, size, cost and paths, with no API call; exit 1 and an auth ✗ line when auth is not ready |
genimg history --json |
past generations: prompt, model, paths, cost |
genimg grid *.png |
an HTML page for a human to pick from |
~/.genimg/metadata/<id>.json |
one generation: absolute paths, prompt, parameters, per-image deltas, cost |
Exit code 2 does not prove a request was sent; see exit codes.