New Example: context-management #10
Description
Context management is a critical but invisible challenge in AI agents. Developers hit these issues in production:
- System prompts silently truncated after long conversations
- Agents lose tool access when context fills up
- Memory retrieval consumes context needed for conversation
- No warning when things break
After simple-agent-with-memory, before react-agent
What it should demonstrates:
- Context Budget Visualization - Real-time token tracking (system/tools/history/available)
- Silent Truncation Problem - Agent forgets instructions mid-conversation, no error thrown
- Memory vs Context Trade-off - Retrieved memories eat context space
- Tool Definition Overflow - 10+ tools consume massive context
- Recovery Strategies - Summarization, sliding windows, graceful resets
Add CODE.md and CONCEPT.md
Real world relevance:
- Explains why chatbots forget personality after 50 messages
- Shows why agents lose tools in long conversations
- Prevents production surprises from token limits
- Improves reliability of deployed AI systems
- Helps developers design around model limits to avoid cut-off responses or lost context