[Skill] summarize-doc command for LLM-friendly document cleanup

[anderskev]

Technology or framework

Document summarization/cleanup for LLM agent consumption.

Why is this needed?

Long-form documentation (whitepapers, guides, API docs) often contains noise that wastes LLM context:

• Author credits, acknowledgements, ToCs
• Marketing language, vendor-specific branding
• Redundant explanations, verbose walkthroughs
• Image placeholders, figure references

A command to distill these to actionable knowledge would make documents usable as skill references or agent context.

Key knowledge areas

1. What to remove (credits, ToC, marketing, vendor names, verbose examples, citations, redundancy)
2. What to retain (core concepts, architectural patterns, best practices, design patterns, security, evaluation approaches, code snippets)
3. Formatting standards (hierarchical headers, bullet points, bold key terms, self-contained sections)
4. Target 80-90% reduction while preserving actionable knowledge

Official documentation

N/A - this is a utility command.

Example use cases

1. Condensing a 50-page whitepaper on AI agent architecture into a ~5-page skill reference
2. Preparing vendor documentation (removing "Google Cloud", "AWS" branding) for generic agent design guidance
3. Creating concise summaries of research papers for context-limited LLM interactions

Would you contribute this skill?

• Yes, I can write this skill
• Yes, I can help review
• No, just requesting

Command Prompt

Clean up this document for use by LLM coding agents designing and implementing AI agents.

**Remove:**
- Author credits, acknowledgements, contributor lists
- Table of contents
- Image placeholders (<!-- image -->), figure references without content
- Marketing/promotional language
- Vendor-specific product names (keep the concepts, remove "Google Cloud", "Vertex AI", etc.)
- Verbose examples and walkthroughs (distill to the pattern)
- Endnotes, references, citations
- Redundant explanations that repeat the same concept
- Enterprise infrastructure specifics not relevant to agent design

**Retain (reorganize for clarity):**
- Core concepts and definitions
- Architectural patterns and components
- Implementation principles and best practices
- Design patterns with clear when-to-use guidance
- Security considerations
- Evaluation and debugging approaches
- Code snippets or pseudocode that illustrate patterns

**Format:**
- Use hierarchical markdown headers
- Bullet points for lists of options/choices
- Bold for key terms on first use
- Keep sections self-contained and scannable
- Target ~80-90% reduction in length while preserving all actionable knowledge

[coderabbitai]

📝 CodeRabbit Plan Mode

Generate an implementation plan and prompts that you can use with your favorite coding agent.

• Create Plan

Examples

• Example 1
• Example 2

---

🔗 Similar Issues

Possible Duplicates

• feat: add summarize-doc command for LLM-friendly document cleanup #26

Related Issues

• feat: Add LLM artifacts review and fix commands #19
• feat: add draft-docs command with Mintlify skills #4
• feat: add write-plan skill optimized for Claude Code #8

🔗 Related PRs

#10 - feat(deepagents): add architecture, implementation, and code review skills [merged]
#11 - fix(commands): add explicit Skill tool instructions for loading skills [merged]
#15 - feat(adr): add ADR skills and write-adr command [merged]
#20 - feat(commands): add LLM artifacts review and fix commands [merged]
#24 - feat(llm-judge): add LLM-as-judge comparison command [closed]

👤 Suggested Assignees

• anderskev
• hey-amelia

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord!

[anderskev]

Additional Requirement: Interactive Product/Marketing Cleanup

Based on real-world usage, the skill should prompt the user whether they want to remove product-specific marketing content and jargon.

Rationale

Some documents may be intentionally vendor-specific (e.g., internal guides for a specific platform), while others benefit from being made vendor-agnostic for broader applicability.

Proposed UX Flow

1. Skill analyzes document and detects vendor-specific content (product names, marketing language, promotional claims)
2. If detected, prompt user with options:
   • Remove vendor-specific content (recommended for generic reference material)
   • Keep vendor-specific content (for platform-specific guides)
   • Review and select (show detected items and let user choose)

Example Prompt

Detected vendor-specific content:
- Product names: Google Agentspace, Vertex AI, NotebookLM (47 mentions)
- Marketing language: "enterprise-grade", "Google quality", "unparalleled" (12 instances)

Would you like to:
1. Remove all vendor-specific content (make document vendor-agnostic)
2. Keep vendor-specific content as-is
3. Review detected items individually

Implementation Notes

• Detection should cover: product names, cloud provider branding, marketing superlatives, enterprise tier descriptions
• Replacement should preserve the underlying concept (e.g., "Vertex AI Search" → "vector search service")
• User preference could be stored for batch processing multiple documents