feat(chat): add agent orchestration system with plan_and_execute

Introduces TaskPlanner and AgentSpec so Pyra can decompose multi-step
tasks into sequential steps, each executed with a focused sub-agent
context rather than the full conversation history.

- plugins/base.py: AgentSpec dataclass + agent_spec() on Protocol/BasePlugin
- plugins/registry.py: register_builtin, get_agent, list_agents
- chat/planner.py: TaskPlanner with plan approval, per-step tool-use loop,
  verification call, and agent-aware routing
- chat/session.py: wires plan_and_execute as a built-in tool after load_all
- chat/history.py: planning hint in system prompt + dynamic agents listing

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

src/pyra/chat/session.py

@@ -14,9 +14,11 @@ from pyra.chat.renderer import (
     render_system,
     render_text_response,
 )
+from pyra.chat.planner import TaskPlanner
 from pyra.config.manager import load_config
 from pyra.config.schema import PyraConfig
 from pyra.memory.reader import list_memories
+from pyra.plugins.base import Tool
 from pyra.plugins.executor import ToolExecutor
 from pyra.plugins.registry import PluginRegistry
 from pyra.security.injection import scan_response
@@ -44,6 +46,37 @@ def start_chat() -> None:
     registry = PluginRegistry.instance()
     registry.load_all(pyra_home() / "plugins", cfg.plugins.enabled)
     executor = ToolExecutor(registry, console)
+    planner = TaskPlanner(cfg, registry, executor)
+    registry.register_builtin(Tool(
+        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. "
+            "Specify 'agent' per step to route to a specialized agent."
+        ),
+        parameters={
+            "type": "object",
+            "properties": {
+                "task": {"type": "string", "description": "Overall task description."},
+                "steps": {
+                    "type": "array",
+                    "items": {
+                        "type": "object",
+                        "properties": {
+                            "description": {"type": "string", "description": "What this step should accomplish."},
+                            "agent": {"type": "string", "description": "Optional agent name to handle this step."},
+                        },
+                        "required": ["description"],
+                    },
+                    "minItems": 1,
+                    "description": "Ordered steps. Each step optionally routes to a named agent.",
+                },
+            },
+            "required": ["task", "steps"],
+        },
+        handler=planner.make_tool_handler(),
+        requires_approval=False,
+    ))
 
     history = ConversationHistory(cfg, registry)
     session: PromptSession = PromptSession(