image-generatelisted
Install: claude install-skill nolte/claude-shared
# Image Generate
Turns a text prompt into an image file on disk — no chat UI, scriptable into any pipeline — by driving the bundled, stdlib-only `scripts/image_generate.py`. Backends are **swappable via `--provider`** so the capability is not locked to one vendor's pricing or availability.
Implements `${CLAUDE_PLUGIN_ROOT}/spec/tools/image-generation/<canonical_language>.md` (per-provider details in `spec/design/flux-image-generation/` and `spec/design/gemini-image-generation/`). Where this skill and the spec disagree, the spec wins.
## Why this is a skill, not an agent
- **Operator-invoked slash command.** Reached as `/nolte-media:image-generate` with a prompt, a provider, and a target path; the operator drives it directly rather than a parent dispatching a fire-and-forget worker.
- **Mid-flow confirmation is part of the contract.** Some providers surface a one-time data/licence notice the operator must acknowledge, and an existing target file must not be overwritten without confirmation. Those are interactive gates an agent's structured-report shape can't carry.
- **The result flows back into the conversation.** Written image and sidecar paths land in the operator's context so the next step (vectorising via `png-to-transparent-svg`, embedding) can follow inline.
- Counter-dimension: the generation itself is a single deterministic script call; that engine is isolated in the bundled script, while the load-bearing dimensions (operator invocation, the acknowledgement/overwr