Context

Comprehensive end-to-end test validating the full single-agent pipeline works correctly. This is the capstone test for M3, ensuring all components integrate properly.

Acceptance Criteria

• Scenario 1: Agent with file tools creates a file from a task description
• Scenario 2: Agent without tools answers a question (text-only response)
• Scenario 3: Tool permission denied is handled gracefully (clear error, no crash)
• Scenario 4: Max iterations reached results in clean failure with informative message
• Mocked LLM provider used (no real API calls in CI)
• Happy path and error paths both covered
• Cost tracking validated: costs recorded correctly for each scenario
• Status transitions validated: correct lifecycle states observed
• Optional real LLM flag for manual integration testing runs
• Tests are deterministic and reproducible

Dependencies

• Depends on Design and implement basic tool system (registry, invocation, results) #15 (single-task execution lifecycle)

Design Spec Reference

Section 3.1, 6.1, 11.1 — Agent System, Task Execution, and Tool System