← ClaudeAtlas

generatelisted

This skill should be used when the user asks to "生成图片", "文生图", "图生图", "根据这张图修改", "创建游戏素材", "生成前端图片资源", "批量生成素材", or invokes `/ekko-image-gen:generate`. It plans context-aware output locations, accepts images pasted into the current Claude Code message, coordinates parallel image-worker agents, reviews generated files, and returns local output paths with host-dependent convenience links.
ZaunEkko/claude-plugins · ★ 0 · AI & Automation · score 78
Install: claude install-skill ZaunEkko/claude-plugins
# Generate images with an OpenAI-compatible service Treat `$ARGUMENTS` and the full current user message as the image brief. Include images pasted or attached to that same message as reference inputs. Route text-only jobs to text-to-image and jobs with references to image editing through the bundled runner. Keep one user-facing command. Do not ask the user to choose between separate text-to-image and image-to-image commands. ## Resolve the task context 1. Read the current task, relevant prior conversation, current working directory, and repository guidance. 2. Inspect `CLAUDE.md`, `AGENTS.md`, existing asset directories, imports, naming patterns, and the code or content that will consume the image when those signals are relevant. 3. Treat an explicit user path or target file as authoritative. 4. Infer the most natural project destination when no path is explicit. Consult `references/output-placement.md` for the decision hierarchy. 5. Fall back to `<project-root>/generated-images/` only when the project provides no reliable destination signal. State the fallback clearly. Perform this inference in the current parent agent. Do not delegate an ambiguous project-wide destination decision to image workers. ## Resolve reference images - Detect images attached or pasted into the current user message and use their host-provided local cache paths directly. - Accept additional local paths, `file:///` links, HTTP(S) URLs, and multiple references. - Upload references as multipart f