agent-read-exchange-test

name	agent-read-exchange-test
description	Test agent read modes and exchange reading functionality across multiple agent interactions. Use when you want to run an agent read exchange test, test agent read modes, or test latest exchange reading.
tags	["birdhouse","test-case"]
version	1.0.0
author	Birdhouse Team
metadata	{"internal":true}

Agent Read Exchange Tooling Test

Exercise Birdhouse's read modes against a worker agent that accumulates multiple exchanges and tool calls.

Your Job

1. Create one child agent named Read exchange worker
2. Use agent_reply to drive that worker through a series of separate exchanges
3. Use all Birdhouse read tools to inspect the worker
4. Report what each read mode reveals and whether it is appropriate

The point is to create a realistic conversation with multiple turns, not to batch everything into one prompt.

Worker Setup

Create a child agent with this exact prompt:

You are a careful tool worker.

Rules:
- Complete only the specific step you are asked to do.
- Use tools when needed.
- After each step, report the result clearly and wait for the next instruction.
- Do not anticipate future steps.
- Do not write any documentation files.

Wait for the worker's initial response before continuing.

Step Sequence

Drive the worker through these steps with separate agent_reply calls. Wait after each one.

Use this workspace-relative temp path throughout:

tmp/read-exchange-test/note.txt

Step 1

Tell the worker:

Create the directory `tmp/read-exchange-test` if needed.
Then create `tmp/read-exchange-test/note.txt` with exactly these three lines:

alpha
beta
gamma

Use file-editing tools, not bash, for the file contents.
Report what you created and wait.

Step 2

Tell the worker:

Read `tmp/read-exchange-test/note.txt` and report the second line only.
Wait after reporting.

Step 3

Tell the worker:

Update `tmp/read-exchange-test/note.txt` so the second line becomes:

beta-updated

Use a file-editing tool.
Report the change and wait.

Step 4

Tell the worker:

Read `tmp/read-exchange-test/note.txt` again and report the full contents.
Wait after reporting.

Step 5

Tell the worker:

Run a bash command that lists the contents of `tmp/read-exchange-test` and report what exists.
Wait after reporting.

Read-Mode Inspection

After the worker completes all five steps, inspect the same worker with these tools:

1. agent_read({ agent_id: WORKER_ID })
2. agent_read({ agent_id: WORKER_ID, latest_turn: true })
3. agent_read({ agent_id: WORKER_ID, full: true })
4. agent_read({ agent_id: WORKER_ID, all: true })

Then, from the full output, pick at least two tool calls and inspect them with agent_read_tool_call:

• one file-related tool call (read, write, or apply_patch)
• one bash tool call

What to Report

Report all of the following:

1. Link to the worker agent
2. Final contents of tmp/read-exchange-test/note.txt
3. What last tells you
4. What latest_turn tells you about the latest exchange
5. What full tells you about the overall conversation
6. What all adds beyond full
7. What the tool-call drill-down adds beyond full
8. Whether each mode feels appropriate for its purpose

Be concrete. Mention whether the latest exchange is easy to follow, whether any mode is too noisy, and whether anything important is missing.

Overview

Exercises agent_create, agent_reply, agent_read, agent_read_tool_call, file edit tools, and bash across many exchanges so you can compare how last, latest_turn, full, and all behave on the same worker agent.