{
  "id": "page:docs-reference-repos-clawhub-steipete-agent-scripts-research",
  "_kind": "Page",
  "_file": "wiki/docs/reference-repos/clawhub/steipete/agent-scripts/research.md",
  "_cluster": "wiki",
  "attributes": {
    "nodeKind": "Page",
    "sourcePath": "docs/reference-repos/clawhub/steipete/agent-scripts/research.md",
    "sourceKind": "repo-docs",
    "title": "steipete/agent-scripts",
    "displayName": "steipete/agent-scripts",
    "slug": "docs/reference-repos/clawhub/steipete/agent-scripts/research",
    "articlePath": "wiki/docs/reference-repos/clawhub/steipete/agent-scripts/research.md",
    "article": "\n# steipete/agent-scripts\n\n- **Archetype**: mega-skill-pack\n- **Stars**: 2,383\n- **Last pushed**: 2026-03-13\n- **License**: NOASSERTION\n- **Discovered**: 2026-04-12\n- **Source**: ClawHub skills (multiple skills published from this repo)\n- **Skills found**: 19 skill directories in skills/\n- **Fork**: No\n\n## Summary\n\nsteipete's shared agent scripts repository containing 19 skills used across his projects and published individually to ClawHub. This is the source repo for several popular ClawHub skills including nano-banana-pro, oracle, brave-search, and others.\n\nSkills found:\n1. **1password** - 1Password CLI integration\n2. **brave-search** - Web search via Brave Search API\n3. **create-cli** - CLI creation tooling\n4. **discord-clawd** - Discord integration\n5. **domain-dns-ops** - Domain/DNS operations\n6. **frontend-design** - Frontend design assistance\n7. **instruments-profiling** - macOS Instruments profiling\n8. **markdown-converter** - Markdown conversion\n9. **nano-banana-pro** - Image generation/editing via Gemini 3 Pro Image\n10. **native-app-performance** - Native app performance analysis\n11. **openai-image-gen** - OpenAI image generation\n12. **openclaw-relay** - OpenClaw relay/orchestration\n13. **oracle** - Cross-model review via GPT-5 Pro\n14. **swift-concurrency-expert** - Swift concurrency guidance\n15. **swiftui-liquid-glass** - SwiftUI Liquid Glass design\n16. **swiftui-performance-audit** - SwiftUI performance auditing\n17. **swiftui-view-refactor** - SwiftUI view refactoring\n18. **video-transcript-downloader** - Video transcript extraction\n19. **xurl** - URL expansion/resolution\n\nThe repo is macOS/Apple-ecosystem heavy (Swift, SwiftUI, Instruments) with cross-cutting utility skills (brave-search, oracle, nano-banana-pro, markdown-converter).\n\n## Assessment\n\nMEDIUM extractable value. Many skills are Apple-ecosystem-specific (SwiftUI, Instruments, native app). The cross-cutting utilities (brave-search, oracle, nano-banana-pro, video-transcript-downloader) have broader value. The oracle skill's cross-model validation pattern is notable -- using one model to review another's work.\n\n**Extraction priority**: MEDIUM\n\n# Extractable Value: steipete/agent-scripts\n\n## Processes\n\n### 1. Cross-Model Code Review\n- **Source**: oracle skill (bundle prompt + files, send to GPT-5 Pro for second opinion)\n- **Placement**: `specializations/shared/cross-model-code-review.js`\n- **Description**: Process for getting a second AI model's review on code changes: select relevant files -> prepare context bundle -> submit to secondary model (API or browser) -> compare responses -> surface disagreements and insights -> create actionable review notes. Uses breakpoints to present findings to the user.\n\n### 2. AI Image Generation Pipeline\n- **Source**: nano-banana-pro skill (draft -> iterate -> final workflow)\n- **Placement**: `specializations/shared/ai-image-generation-pipeline.js`\n- **Description**: Multi-step image generation process: draft at 1K resolution -> review with user -> iterate on prompt adjustments -> upscale to 4K when prompt is locked. Supports both text-to-image and image-to-image editing. Breakpoints for user review at each draft iteration.\n\n### 3. Video Content Summarization\n- **Source**: video-transcript-downloader + summarize patterns\n- **Placement**: `specializations/shared/video-content-summarization.js`\n- **Description**: Process for extracting and summarizing video content: download transcript (YouTube/direct URL) -> extract key segments -> generate structured summary -> identify action items and key quotes -> export as Markdown.\n\n## Plugin Ideas\n\n### 1. Web Search Plugin (Brave)\n- **Category**: Tools Integration\n- **install.md**: Installs brave-search skill dependencies (node), configures BRAVE_API_KEY, provides babysitter tasks for web search (query with result count) and page content extraction (URL to markdown). Lightweight alternative to browser automation for information retrieval.\n\n### 2. Cross-Model Validation Plugin (Oracle)\n- **Category**: QA & Testing\n- **install.md**: Installs oracle CLI (@steipete/oracle), configures secondary model access (OpenAI API key or browser automation). Provides a babysitter task that bundles process context + files and gets a second model's opinion. Useful as a quality gate in any process -- the \"phone a friend\" pattern.\n\n### 3. Image Asset Generation Plugin\n- **Category**: Tools Integration\n- **install.md**: Installs nano-banana-pro dependencies (uv, Python), configures GEMINI_API_KEY. Provides babysitter tasks for image generation (text-to-image) and image editing (image-to-image) at configurable resolutions. Useful for processes that need to generate visual assets (documentation, marketing, UI mockups).\n\n## Library Mapping\n\n| Extractable Process | Library Status | Action | Existing Path | Target Placement |\n|-------------------|----------------|--------|---------------|------------------|\n| Cross-Model Code Review | NEW | Second AI model review for code changes with disagreement surfacing and actionable notes | - | specializations/shared/cross-model-code-review.js |\n| AI Image Generation Pipeline | NEW | Multi-step image generation: draft → iterate → final with resolution progression | - | specializations/shared/ai-image-generation-pipeline.js |\n| Video Content Summarization | NEW | Video transcript extraction and summarization with key segments and action items | - | specializations/shared/video-content-summarization.js |\n| Draft-Iterate-Final Resolution Pattern | NEW | Low-resolution drafts for iteration, upscale only when locked (for expensive operations) | - | specializations/shared/draft-iterate-final-resolution.js |\n\n## Plugin Marketplace Mapping\n\n| Plugin Idea | Marketplace Status | Action | Existing Plugin | Target Placement |\n|-------------|-------------------|--------|-----------------|------------------|\n| Web Search Integration (Brave) | NEW | Brave Search API integration for web search and page content extraction | - | plugins/a5c/marketplace/plugins/web-search-integration/ |\n| Cross-Model Validation Oracle | NEW | Second model opinion as quality gate with adversarial review pattern | - | plugins/a5c/marketplace/plugins/cross-model-validation-oracle/ |\n| Image Asset Generation | NEW | Gemini-powered image generation and editing with configurable resolutions | - | plugins/a5c/marketplace/plugins/image-asset-generation/ |\n\n## Implicit Procedural Knowledge\n\n- **Draft-iterate-final resolution pattern**: nano-banana-pro's workflow of starting at low resolution for fast iteration then upscaling only when prompt is locked. This pattern applies to any expensive generative operation -- do cheap drafts first.\n- **Cross-model review as quality gate**: oracle demonstrates using a different model (GPT-5 Pro) to review work done by the primary model. This \"adversarial review\" pattern is a powerful quality mechanism for babysitter processes.\n- **Skill as thin CLI wrapper**: Many skills here are thin SKILL.md wrappers around CLI tools (brave-search, oracle). The pattern of SKILL.md -> bash script -> CLI tool is the most common skill architecture on ClawHub.\n",
    "documents": []
  },
  "outgoingEdges": [],
  "incomingEdges": [
    {
      "from": "page:docs-reference-repos",
      "to": "page:docs-reference-repos-clawhub-steipete-agent-scripts-research",
      "kind": "contains_page"
    }
  ]
}