agent-read-exchange-test

Tests Birdhouse agent read modes across multi-turn worker agent exchanges.

36|5|Updated Feb 27, 2026

Search Tags:#agent testing #multi-turn #test case #birdhouse #read modes #agent_read #tool inspection

Ready to run

npx skills add https://github.com/Birdhouse-Labs/birdhouse --skill agent-read-exchange-test-birdhouse-labs

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: agent-read-exchange-test
Source: https://github.com/Birdhouse-Labs/birdhouse/tree/main/.agents/skills/internal/test-cases/agent-read-exchange-test
Command: npx skills add https://github.com/Birdhouse-Labs/birdhouse --skill agent-read-exchange-test-birdhouse-labs

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Verifying that Birdhouse's agent inspection tools behave correctly requires a realistic multi-turn conversation, which is tedious to construct manually. This Skill automates the creation of a worker agent with multiple exchanges and tool calls so read modes can be compared against the same conversation. ## Core Features & Use Cases - Multi-Exchange Worker Setup: Creates a child worker agent and drives it through five separate file-editing and bash steps using agent_reply. - Read-Mode Comparison: Inspects the same worker with agent_read in last, latest_turn, full, and all modes to compare what each reveals. - Tool-Call Drill-Down: Uses agent_read_tool_call on file and bash tool calls to verify granular inspection beyond the full transcript. - Use Case: A Birdhouse developer changes the agent_read implementation and runs this test to confirm each read mode returns appropriate, non-noisy output for a realistic conversation. ## Quick Start Run the agent read exchange test to create a worker agent, drive it through five steps, and compare all agent_read modes.

Frequently Asked Questions about agent-read-exchange-test

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I test Birdhouse agent read modes?▼

Create a child worker agent, drive it through several separate exchanges with agent_reply, then call agent_read with last, latest_turn, full, and all options. Comparing the outputs on the same worker shows what each mode reveals.

What is the difference between agent_read full and all modes?▼

The full mode returns the overall conversation history for an agent, while the all mode adds additional detail beyond full. This test inspects the same worker with both modes so you can report exactly what all adds.

How do I inspect a specific tool call in a Birdhouse agent?▼

Use agent_read_tool_call with the tool call identifier taken from the agent_read full output. This drills into individual calls such as file edits or bash commands beyond what the transcript summary shows.

Why should test exchanges use separate agent_reply calls?▼

Separate agent_reply calls create a realistic multi-turn conversation with distinct exchanges and tool calls. Batching everything into one prompt would not exercise how read modes handle turn boundaries and accumulated history.

When should I use latest_turn instead of full in agent_read?▼

Use latest_turn when you only need the most recent exchange without the noise of the full history. Use full when you need the overall conversation context across all exchanges and tool calls.