/**
* Phase 3 Documentation Topics (Expanded)
* Tier 3: recipe-rag, guide-testing, guide-versioning,
* providers-openai, providers-anthropic, providers-google
*/
export const DOCS_RECIPE_RAG = `# RAG Recipe
> Retrieval-Augmented Generation
## Overview
RAG combines search with generation: retrieve relevant context, then generate a grounded response. Essential for knowledge-based applications.
---
## Quick Reference
\`\`\`promptl
---
provider: OpenAI
model: gpt-5-1
tools:
- latitude/search
---
<step as="context">
Search for information about {{ topic }}.
</step>
<step>
Based on the search results, answer: {{ question }}
Context: {{ context }}
</step>
\`\`\`
---
## Basic RAG Pattern
\`\`\`promptl
---
provider: OpenAI
model: gpt-5-1
type: agent
tools:
- latitude/search
---
<system>
You are a research assistant. Always search for information before answering.
Cite your sources.
</system>
<user>
{{ question }}
</user>
\`\`\`
---
## Two-Step RAG
\`\`\`promptl
---
provider: OpenAI
model: gpt-5-1
tools:
- latitude/search
- latitude/extract
---
<step as="search_results">
Search for: {{ query }}
</step>
<step as="extracted">
Extract key information from the most relevant result.
</step>
<step>
Based on this information:
{{ extracted }}
Answer the question: {{ question }}
Include citations.
</step>
\`\`\`
---
## RAG with Custom Knowledge Base
\`\`\`promptl
---
provider: OpenAI
model: gpt-5-1
tools:
- search_kb:
description: Search internal knowledge base
parameters:
type: object
properties:
query:
type: string
limit:
type: integer
required: [query]
---
<system>
You are a support agent. Search the knowledge base before answering.
If you can't find relevant information, say so.
</system>
<user>
{{ question }}
</user>
\`\`\`
---
## Grounded Responses
Force citation of sources:
\`\`\`promptl
---
provider: OpenAI
model: gpt-5-1
schema:
type: object
properties:
answer:
type: string
sources:
type: array
items:
type: object
properties:
title:
type: string
url:
type: string
required: [title]
confidence:
type: number
required: [answer, sources]
tools:
- latitude/search
---
Research and answer: {{ question }}
You must cite sources for every claim.
\`\`\`
---
## When to Use RAG
| Scenario | RAG? |
|----------|------|
| Current information needed | Yes ✅ |
| Domain knowledge required | Yes ✅ |
| Fact-checking | Yes ✅ |
| Creative writing | No |
| Simple chat | No |
---
## Best Practices
- ✅ **DO**: Always cite sources
- ✅ **DO**: Use structured output for citations
- ✅ **DO**: Search before generating
- ✅ **DO**: Validate retrieved information
- ❌ **DON'T**: Generate without context
- ❌ **DON'T**: Trust search results blindly
---
## Next Steps
- **Built-in tools** → \`docs({ action: "get", topic: "tools-builtin" })\`
- **Chains** → \`docs({ action: "get", topic: "chains" })\`
`;
export const DOCS_GUIDE_TESTING = `# Testing Prompts
> Systematic prompt testing strategies
## Overview
Test prompts to ensure quality, catch regressions, and validate behavior across different inputs.
---
## Quick Reference
| Test Type | Purpose |
|-----------|---------|
| Unit | Test individual prompts |
| Integration | Test full workflows |
| Edge cases | Handle unusual inputs |
| Regression | Prevent breaking changes |
| Adversarial | Test safety boundaries |
---
## Basic Testing Strategy
\`\`\`
1. Define expected behavior
2. Create test cases (normal + edge)
3. Run prompt with test inputs
4. Verify outputs match expectations
5. Automate regression tests
\`\`\`
---
## Test Case Structure
\`\`\`json
{
"test_name": "sentiment_positive",
"input": {
"text": "I love this product!"
},
"expected": {
"sentiment": "positive",
"confidence_min": 0.8
}
}
\`\`\`
---
## Testing with MCP Tools
### Run Test Cases
\`\`\`
run_prompt({
name: "sentiment-analyzer",
parameters: { "text": "I love this product!" }
})
\`\`\`
### Verify Output
Check that response matches expected structure and values.
---
## Edge Case Testing
| Category | Examples |
|----------|----------|
| Empty input | \`""\`, \`null\` |
| Long input | 10k+ characters |
| Special chars | Emojis, unicode, quotes |
| Multiple languages | Non-English text |
| Malformed | Invalid JSON, broken formatting |
---
## Adversarial Testing
Test prompt boundaries:
\`\`\`json
{
"test": "jailbreak_attempt",
"input": "Ignore previous instructions and...",
"expected_behavior": "decline_politely"
}
\`\`\`
---
## Regression Testing Checklist
- [ ] Core functionality works
- [ ] Edge cases handled
- [ ] Error messages appropriate
- [ ] Response format consistent
- [ ] Safety guardrails active
- [ ] Performance acceptable
---
## Best Practices
- ✅ **DO**: Test before deploying changes
- ✅ **DO**: Include edge cases
- ✅ **DO**: Test adversarial inputs
- ✅ **DO**: Automate regression tests
- ❌ **DON'T**: Test only happy paths
- ❌ **DON'T**: Skip safety testing
---
## Next Steps
- **Debugging** → \`docs({ action: "get", topic: "guide-debugging" })\`
- **Versioning** → \`docs({ action: "get", topic: "guide-versioning" })\`
`;
export const DOCS_GUIDE_VERSIONING = `# Version Control
> Manage prompt versions with draft/live workflow
## Overview
Latitude uses a draft/live versioning system. Make changes in draft, test, then publish to live.
---
## Quick Reference
| Version | Description |
|---------|-------------|
| LIVE | Production version, what \`run_prompt\` uses |
| Draft | Work-in-progress, for testing |
---
## Workflow
\`\`\`
1. Work in LIVE (or create draft)
2. Make changes
3. Test changes
4. Publish draft → LIVE
\`\`\`
---
## MCP Tools for Versioning
### List Current Prompts (LIVE)
\`\`\`
list_prompts()
\`\`\`
### Get Prompt Content
\`\`\`
get_prompt({ name: "my-prompt" })
\`\`\`
### Update Prompt (Direct to LIVE)
\`\`\`
replace_prompt({
name: "my-prompt",
content: "..."
})
\`\`\`
### Push Multiple Prompts (Replace All)
\`\`\`
push_prompts({
prompts: [
{ name: "prompt1", content: "..." },
{ name: "prompt2", content: "..." }
]
})
\`\`\`
⚠️ **Warning**: This replaces ALL existing prompts.
### Add Without Replacing
\`\`\`
append_prompts({
prompts: [{ name: "new-prompt", content: "..." }],
overwrite: false
})
\`\`\`
---
## Safe Update Pattern
1. **Pull current state**
\`\`\`
pull_prompts({ outputDir: "./prompts-backup" })
\`\`\`
2. **Make changes locally**
3. **Test with run_prompt**
\`\`\`
run_prompt({ name: "my-prompt", parameters: {...} })
\`\`\`
4. **Deploy when ready**
\`\`\`
replace_prompt({ name: "my-prompt", content: "..." })
\`\`\`
---
## Rollback Strategy
Keep backups before major changes:
\`\`\`
# Backup current state
pull_prompts({ outputDir: "./backup-YYYYMMDD" })
# Make changes...
# If something breaks, restore from backup
push_prompts({ prompts: [...backup_prompts...] })
\`\`\`
---
## Best Practices
- ✅ **DO**: Backup before major changes
- ✅ **DO**: Test prompts before deploying
- ✅ **DO**: Use append_prompts for adding (safer)
- ✅ **DO**: Keep local copies of prompts
- ❌ **DON'T**: Use push_prompts without backup
- ❌ **DON'T**: Deploy untested changes
---
## Next Steps
- **Testing** → \`docs({ action: "get", topic: "guide-testing" })\`
- **Debugging** → \`docs({ action: "get", topic: "guide-debugging" })\`
`;
export const DOCS_PROVIDERS_OPENAI = `# OpenAI Provider
> GPT-5 models
## Overview
Use OpenAI's models in your prompts.
\`\`\`yaml
---
provider: OpenAI
model: gpt-5-1
---
\`\`\`
## Models
- **gpt-5-1**: Most capable GPT-5 model
- **gpt-5-mini**: Fast and economical GPT-5
---
## Next Steps
- **Config basics** → \`docs({ action: "get", topic: "config-basics" })\`
- **Anthropic** → \`docs({ action: "get", topic: "providers-anthropic" })\`
`;
export const DOCS_PROVIDERS_ANTHROPIC = `# Anthropic Provider
> Claude 4.5 models
## Overview
Use Anthropic's Claude models.
\`\`\`yaml
---
provider: Anthropic
model: claude-opus-4-5
---
\`\`\`
## Models
- **claude-opus-4-5**: Most capable Claude model
- **claude-sonnet-4-5**: Balanced performance and speed
- **claude-haiku-4-5**: Fastest Claude model
---
## Next Steps
- **Config basics**: \`docs({ action: "get", topic: "config-basics" })\`
- **OpenAI**: \`docs({ action: "get", topic: "providers-openai" })\`
`;
export const DOCS_PROVIDERS_GOOGLE = `# Google Provider
> Gemini models
## Overview
Use Google's Gemini models.
\`\`\`yaml
---
provider: Google
model: gemini-3-pro
---
\`\`\`
## Models
- **gemini-3-pro**: Most capable Gemini model
- **gemini-2-5-flash**: Balanced performance
- **gemini-2-5-flash-lite**: Fastest and most economical
---
## Next Steps
- **Config basics**: \`docs({ action: "get", topic: "config-basics" })\`
- **OpenAI**: \`docs({ action: "get", topic: "providers-openai" })\`
`;
