docs(agent): record performance fix notes

Author: pythoninthegrass

.serena/memories/agent_performance_fixes.md

@@ -0,0 +1,17 @@
+# Agent Performance Fixes (TASK-277)
+
+## Issues
+1. **Parallel tool execution not working** - Model not calling multiple tools per turn
+2. **Token cap needed** - Response generation taking 63s due to excessive token generation
+
+## Changes Required
+
+### 1. Add max_tokens to agent builder (mod.rs)
+Add `.max_tokens(1024)` after `.temperature(0.2)` in the `build_agent()` function.
+
+### 2. Update system prompt (prompt.rs)
+Add explicit instruction to call multiple tools in parallel. The prompt should say something like:
+"You can and SHOULD call multiple tools in a single turn. Use parallel tool calls to gather information faster."
+
+### 3. Run tests
+- `cargo nextest run --workspace --features agent` should pass