菜单
Generate slide visuals using Gemini Pro from validated graphics descriptions. Use when you want to generate slide images, create visuals after validation, or see what the slides will look like.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Manage a local task/backlog list (tasks.json + a generated TASKS.md) for the current repo and reconcile it with GitHub or Gitea issues, via the bundled task-sync Python tool. Supports direct commands (list/add/edit/done/remove/ status/init) and a plan-decide-apply sync that surfaces creates, pushes, pulls, conflicts, and confidentiality findings for review before anything leaves the machine, with a public-repo push warning and a --dry-run mode. Suggest when — the user wants to see or manage this repo's tasks or backlog, sync tasks with GitHub/Gitea issues, mentions a local task tracker or TASKS.md, or asks to push/pull/reconcile tasks against issues; keywords — task list, backlog, sync issues, task-sync, todo list, reconcile issues.
Initialize mandatory experiment logging using scientific notebook, ADR, and postmortem patterns. For projects involving system changes, benchmarks, debugging, or exploratory work. Creates LAB_NOTEBOOK.md with living decision/action tracking and injects iron-clad CLAUDE.md rules that make logging a PRECONDITION for every action. Suggest (do not auto-run) when — infrastructure/experimental/expensive-failure projects, multi-session work, or keywords like optimize/benchmark/debug/configure.
Generate BPMN 2.0 compliant XML files from natural language process descriptions OR from structured markdown business process documents. Use this skill when a user wants to create a BPMN workflow, convert a business process to BPMN XML, model a workflow diagram, or generate process definitions. Triggers on requests like "create a BPMN", "generate workflow XML", "model this process", "convert to BPMN 2.0", "create process diagram", "build workflow", or "convert this markdown to BPMN". Do NOT use for converting an existing BPMN XML file to Draw.io — use bpmn-to-drawio for that.
Analyze technical documents for CS/ML concepts a smart non-CS reader wouldn't understand, recommend explanation mechanisms (glossary, inline, footnote, sidebar, appendix), present analysis for approval, then implement annotations in the Word document using the project directory for contextual accuracy. Use when making technical documents accessible to non-CS audiences. Do NOT use for writing a new overview document from a codebase — use explain-project for that.
Comprehensive 9-agent architecture review — spawns parallel domain specialists (architecture, code, data, integration, performance, QA, security, platform, risk) and produces structured findings with executive report and go/no-go recommendation
Set up a persistent, LLM-maintained wiki inside any project. Creates a wiki/ directory with sources, pages, schema, and navigation files, seeds initial pages from project discovery, and injects CLAUDE.md rules that make Claude automatically maintain the wiki during normal work sessions. Suggest (do not auto-run) when — project accumulating complexity, context loss between sessions, multiple contributors, "I keep forgetting", significant domain knowledge, or LAB_NOTEBOOK.md with durable insights.
| name | sg-generate-images |
| description | Generate slide visuals using Gemini Pro from validated graphics descriptions. Use when you want to generate slide images, create visuals after validation, or see what the slides will look like. |
| argument-hint | <presentation.md> [--resolution high|medium|small] |
| allowed-tools | Bash, Read, Glob, Grep |
Generate slide visuals using Gemini Pro from the validated graphics descriptions in the presentation. Uses a unified prompt builder and async batch generation for efficiency.
Working directory:
!pwd
Available presentation files:
!ls -la *.md 2>/dev/null | grep -i pres || echo "No presentation files found"
Existing images:
!ls -la images/ 2>/dev/null || echo "No images directory yet"
sg --version must succeed before proceeding — if sg is missing, stop and tell the user this requires the private davistroy/slide-generator engine (owner-only; see ADR-0008)slide-generator package installedGOOGLE_API_KEY set in environment (Gemini Pro access)presentation.md file with validated graphics descriptions.graphics_validated marker)Required:
<presentation.md> - Path to presentation markdown fileOptional:
--resolution <level> - Image resolution: small, medium, high (default: high)--image-format <fmt> - Output format: png, jpg (default: png)--skip-existing - Don't regenerate images that already existGOOGLE_API_KEY is set and presentation has validated graphicssg generate-images presentation.md --resolution high
images/ directoryThe image generation step uses:
gemini-3-pro-image-preview): Google's image generation modelprompt_builder.py constructs optimized prompts from descriptionsasyncio| Error | Cause | Fix |
|---|---|---|
GOOGLE_API_KEY not found | Missing env var | Set the key |
Image blocked | Content policy violation | Review and soften the description |
Rate limit | Too many concurrent requests | Built-in rate limiter handles this |
Generation timeout | Gemini service slow | Retry; increase SG_API_TIMEOUT |
The step produces:
images/ directory (one per slide)slide-01.png, slide-02.png, etc.)After images are generated, build the final PowerPoint:
sg build presentation.md --template generic
Or use /sg-build to continue the pipeline.
This step delegates image generation to the external sg CLI (the slide-generator package), which is unbundled and maintained outside this repo. personal-plugin's visual-explainer skill/tool is a separate, in-tree, tested Gemini image path with its own prompt builder and model config — the two are intentionally independent implementations, not a shared library. Future Gemini model or API changes (e.g. a new model ID, auth scheme, or block-reason handling) must be applied in both places: here (via the external sg package) and in plugins/personal-plugin/tools/visual-explainer/.