User requested analysis of "Task" tool definition and usage in the SDK. No such tool exists—the repository implements a generic Tool system where tools are identified by string names and configured dynamically.

Documentation Added

• ANALYSIS_README.md - Entry point with bilingual quick reference
• TASK_TOOL_ANALYSIS.md / TASK_TOOL_ANALYSIS_EN.md - Complete analysis of Tool concepts, state lifecycle, and usage patterns
• TOOL_ARCHITECTURE.md - ASCII diagrams showing system architecture, state transitions, and data flow
• examples/tool_usage_example.py - Working code examples for sync/async clients, state monitoring, and error handling

Key Findings

Tool system architecture:

• ToolPart: Execution tracking with states (pending → running → completed/error)
• Configuration: Dict[str, bool] in Mode/Session: {"tool_name": True}
• Response Processing: Tools appear as discriminated union type in AssistantMessage.parts[]

Usage pattern:

from opencode_ai import Opencode
from opencode_ai.types import TextPartInputParam

client = Opencode()
session = client.session.create()

response = client.session.chat(
    id=session.id,
    model_id="...",
    provider_id="...",
    parts=[TextPartInputParam(text="...", type="text")],
    tools={"file_reader": True, "search": False}
)

for part in response.parts:
    if part.type == "tool" and part.state.status == "completed":
        print(f"{part.tool}: {part.state.output}")

Types defined in src/opencode_ai/types/tool_*.py with full state machine implementation across ToolStatePending, ToolStateRunning, ToolStateCompleted, ToolStateError.

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.