agent-patternslisted

# Agent Patterns Five composable workflow patterns from Anthropic's "Building Effective Agents" guide. Pick the simplest pattern that solves the problem. Add complexity only when it demonstrably improves outcomes. ## Decision Framework Before picking a pattern, ask: 1. **Can a single optimized LLM call with retrieval + examples solve this?** If yes, stop here. No agent needed. 2. **Are the subtasks predictable and fixed?** → Use a **workflow** (patterns 1-4) 3. **Are subtasks unpredictable, requiring model-driven decisions?** → Use an **agent** (pattern 5+) ## The 5 Patterns ### 1. Prompt Chaining **What**: Sequential steps, each LLM call processes the previous output. Optional programmatic gates between steps. **When**: Task cleanly decomposes into fixed subtasks. Trade latency for accuracy by making each call easier. **Structure**: ``` Input → LLM₁ → [Gate] → LLM₂ → [Gate] → LLM₃ → Output ``` **Examples**: - Generate copy → translate to target language - Write outline → validate against criteria → write full document - Extract data → transform → generate report **Implementation**: Chain calls with validation checks between steps. If a gate fails, loop back or abort. --- ### 2. Routing **What**: Classify input, direct to specialized handler. Separation of concerns. **When**: Distinct input categories need different prompts/tools/models. Optimizing for one category hurts others. **Structure**: ``` Input → Classifier → Route A (specialized prompt)