Build test harness for Pi coding agent

[bBlazewavE]

Summary

Create a test harness for validating Pi coding agent behavior — ensuring it responds correctly to prompts, handles tool calls, and integrates with the dev session workflow.

Proposed approach

• A tests/ directory with scripted test scenarios for Pi
• Harness should verify: agent startup, prompt/response cycles, tool invocations, error handling
• Support headless execution (no tmux required) for CI compatibility
• Output structured results (pass/fail per test, timing, token usage)

Example scenarios

• Pi launches and responds to a basic prompt
• Pi correctly uses Claude as a provider
• Pi handles missing API keys gracefully
• Pi integrates with MCP servers when configured

Acceptance criteria

• just test-pi recipe runs the harness
• Tests pass in both interactive and headless modes
• Results are human-readable and CI-parseable

[bBlazewavE]

This is a great idea, but creating a robust test harness for Pi requires deeper understanding of the Pi SDK and its testing conventions. I'm deferring this to a later sprint when I can research the Pi documentation thoroughly. For now, prioritizing foundational improvements.