Add LangChain.js integration guide and expand incident documentation

[amavashev]

Summary

This PR expands the Cycles documentation with a comprehensive LangChain.js integration guide and significantly enhances incident documentation with TypeScript examples, severity analysis, detection strategies, monitoring rules, and testing patterns.

Key Changes

New Documentation

• LangChain.js Integration Guide (how-to/integrating-cycles-with-langchain-js.md): Complete guide for adding budget governance to LangChain.js applications using a custom CyclesBudgetHandler callback handler that wraps every LLM call with Cycles reservations. Includes examples for chat models, agents with tools, and streaming responses.

• Glossary (glossary.md): Comprehensive definitions of key Cycles concepts including Budget Authority, Reservation, Commit, Release, Estimate, Decide, Scope, Subject, Three-Way Decision, Overage Policy, and Units.

• Integrations Overview (how-to/integrations-overview.md): High-level overview of all supported Cycles integrations across LLM providers, frameworks, and web servers with language support and streaming capabilities.

• 404 Page (404.md): Custom 404 error page with navigation to key documentation sections.

• Web App Manifest (public/manifest.json): PWA manifest for installable documentation site.

Enhanced Incident Documentation

Concurrent Agent Overspend (incidents/concurrent-agent-overspend.md):

• Added Python and TypeScript code examples demonstrating the problem and solution
• Added severity and impact analysis with concrete cost multiplication examples
• Added detection strategies (querying active reservations, spotting TOCTOU patterns)
• Added comprehensive Prometheus alerting rules
• Added concurrent load testing patterns in both Python and TypeScript

Retry Storms and Idempotency Failures (incidents/retry-storms-and-idempotency-failures.md):

• Added TypeScript equivalent using withCycles wrapper
• Added severity and impact analysis showing cost explosion across retries and fleet multiplication
• Added detection strategies (reservation-to-commit ratio analysis, idempotency key pattern matching)
• Added Prometheus alerting rules for retry storm detection
• Expanded prevention strategies with detailed implementation examples in both languages
• Added testing patterns for verifying retry behavior under budget constraints

Scope Misconfiguration and Budget Leaks (incidents/scope-misconfiguration-and-budget-leaks.md):

• Added severity and impact section explaining silent nature of scope misconfiguration
• Added pattern Add comprehensive deployment guide and admin server documentation #5: dynamic scope values from user input with sanitization examples
• Added TypeScript detection example for programmatically finding scope gaps
• Added Prometheus alerting rules for scope mismatch detection
• Added testing patterns for verifying scope configuration consistency
• Added centralized scope builder pattern with TypeScript implementation

Enhanced How-To Guides

Handling Streaming Responses with Cycles (how-to/handling-streaming-responses-with-cycles.md):

• Added TypeScript section with reserveForStream pattern for streaming responses
• Includes automatic heartbeat handling and cap application examples

Configuration Updates

VitePress Config (.vitepress/config.ts):

• Added apple-touch-icon and manifest.json links for PWA support

Notable Implementation Details

• The LangChain.js callback handler uses runId to track concurrent reservations, ensuring proper lifecycle management for multi-turn agents
• Incident documentation now includes both Python and TypeScript examples consistently across all patterns
• Detection and monitoring sections provide actionable Prometheus queries and alerting rules
• Testing patterns demonstrate how to verify budget enforcement under concurrent and retry scenarios
• Scope configuration examples emphasize input validation and sanitization to prevent bypass vulnerabilities

https://claude.ai/code/session_01LVeAZGvccG11HfdcBGqECM