chore(chat): tighten tool descriptions to reduce AI selection confusion

- plan_and_execute: restrict to 3+ step tasks; prevents over-triggering on simple requests
- memory_read: hint to call memory_lookup first to find the correct path

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

src/pyra/chat/session.py

@@ -81,8 +81,9 @@ def start_chat() -> None:
         name="plan_and_execute",
         description=(
             "Decompose a multi-step task into sequential steps and execute each with "
-            "a focused sub-agent. Use when the request has multiple distinct phases. "
+            "a focused sub-agent. Use only for long tasks with 3 or more sequential "
-            "Specify 'agent' per step to route to a specialized agent."
+            "steps that need verification between them — for simple 2-step tasks, do "
+            "them directly. Specify 'agent' per step to route to a specialized agent."
         ),
         parameters={
             "type": "object",
@@ -125,7 +126,7 @@ def start_chat() -> None:
     ))
     registry.register_builtin(Tool(
         name="memory_read",
-        description="Read the full content of a memory file by its relative path (e.g. 'user/profile.md').",
+        description="Read the full content of a memory file by its relative path (e.g. 'user/profile.md'). Use memory_lookup first to find the correct path.",
         parameters={
             "type": "object",
             "properties": {