Stack
Stack pages describe capability slots, not products. Each slot says what the capability must do, how to select for it, and what to avoid.
- Agent Architecture: decompose, delegate, validate, and recover
- Cost Management: cost per accepted win, tokens plus human attention
- Evaluation & Benchmarking: define the win and measure whether the stack works
- Harness Selection: the interface to the models
- Memory & Context: in-session context, cross-session memory, standing instructions
- Model Selection: tiers, routing policy, and gateway
- Observability & Logging: traces, logs, and metrics for agent runs
- Repository Context: indexed codebase knowledge for agents
- Skills & Prompts: repeatable workflows, encoded
- Tool Integration: tool design, context cost, and security