Claude Code Plugins

Community-maintained marketplace

Feedback

Navigate project documentation efficiently. Use when asked to "read the docs", "check documentation", or before starting unfamiliar tasks. Triggers on read docs, check docs, documentation, unfamiliar codebase.

Install Skill

1Download skill
2Enable skills in Claude

Open claude.ai/settings/capabilities and find the "Skills" section

3Upload to Claude

Click "Upload skill" and select the downloaded ZIP file

Note: Please verify skill by going through its instructions before using it.

SKILL.md

name docs-navigator
description Navigate project documentation efficiently. Use when asked to "read the docs", "check documentation", or before starting unfamiliar tasks. Triggers on read docs, check docs, documentation, unfamiliar codebase.

Documentation Navigator

Efficiently read project documentation without bloating context.

When to Use

  • User explicitly says "read the docs"
  • Starting work in unfamiliar area of codebase
  • Need to understand patterns/conventions before implementing

Workflow

Step 1: Read the Index

Always start with docs/index.md:

  • Contains keywords and summaries for all docs
  • Enables smart routing to relevant docs only

Step 2: Match Task to Keywords

Look at the Keywords column in the index:

  • Identify 1-3 docs that match your current task
  • Don't read docs that don't match

Example matching:

  • Task: "Fix authentication bug" → Match: auth, login, error handling
  • Task: "Add new API endpoint" → Match: REST, routes, architecture
  • Task: "Deploy to production" → Match: deploy, CI/CD, environments

Step 3: Read Only Relevant Docs

  • Read the 1-3 matched docs fully
  • Apply patterns and conventions found there
  • Skip unrelated docs to preserve context

Step 4: Apply What You Learned

  • Follow documented patterns exactly
  • Don't deviate from established conventions
  • Reference specific docs in your reasoning

Anti-Patterns

Don't Do This Do This Instead
Read all docs in index Read only task-relevant docs
Skim docs superficially Read relevant docs thoroughly
Ignore index, guess from memory Always check index first
Read docs once and forget Re-read when switching tasks

Expected docs/index.md Format

Projects using this skill should have a docs/index.md with this structure:

# Documentation Index

## How to Use This Index

1. Scan the **Keywords** column for terms matching your task
2. Read the **Summary** to confirm relevance
3. Read ONLY the docs that match - don't read everything

## Documentation Map

| Doc | Keywords | Summary |
|-----|----------|---------|
| [architecture.md](architecture.md) | system design, components, data flow | High-level system architecture |
| [api/endpoints.md](api/endpoints.md) | REST, routes, handlers, requests | API endpoint specifications |
| [guides/testing.md](guides/testing.md) | tests, pytest, mocking, fixtures | Testing patterns |

## Quick Reference

- **Adding a feature?** → Read architecture.md first
- **Fixing a bug?** → Check api/errors.md and operations/monitoring.md

Integration with Hooks

This skill complements the hooks:

  • SessionStart: Reminds to read docs/index.md
  • read-docs-trigger: When user says "read the docs", this skill guides the process
  • Stop: Validates work against documented standards

Example Session

User: Read the docs before we add a new payment endpoint

Claude: [Reads docs/index.md]

I found these relevant docs based on keywords:
- architecture.md (system design, components)
- api/endpoints.md (REST, routes, handlers)
- guides/authentication.md (auth, permissions)

[Reads those 3 docs]

Based on the documentation:
- New endpoints go in src/api/routes/
- Must follow RESTful conventions from api/endpoints.md
- Payment endpoints require auth middleware per guides/authentication.md
- Error responses must match format in api/errors.md

Ready to implement the payment endpoint following these patterns.

Context Efficiency

This approach optimizes context usage:

Approach Context Cost Effectiveness
Read all docs High (10k+ tokens) Unfocused
Read index only Low (500 tokens) Incomplete
Index + matched docs Medium (2-4k tokens) Targeted

By matching keywords before reading, you get the right context without bloat.