agent_read — Claude Code subagent: features and install
This Claude Code subagent reads a selected agent’s recent conversation. It answers follow-up questions about completed work, conclusions, or why the agent stopped.
MIT ★ 395 updated 2026-11-16
.claude/agents/agent_read.mdWhat agent_read does
agent_read returns recent user messages, the agent’s replies, and brief records of its tool calls. This lets another agent answer questions about work already done without using agent_send to start another turn.
A request needs machine and session_id, either the full session ID or its first 8 characters. Set turns to choose how many recent exchanges to read; it defaults to 2 and has a maximum of 10. A stopped agent accepts no messages, so its conversation is read-only.
For Claude Code, install the subagent file at .claude/agents/agent_read.md.
What it helps with
- retrieve an agent’s conclusion from its recent conversation
- find a PR link in an agent’s history
- learn why an agent stopped
- check which tools an agent called
Example prompts for the agent
- Read the recent conversation for session
abc12345onmachineand tell me why the agent stopped. - Check the last 4 exchanges for session
abc12345and find the PR link. - Read the agent’s history for session
abc12345and report its conclusion.
How to install agent_read
Manually: put the files in place
.claude/agents/agent_read.mdFor all projects use ~/.claude/agents/. Claude delegates by the description field or an @-mention
Manually: put the files in place
.codex/agents/agent_read.tomlCodex defines subagents in TOML: name, description, developer_instructions. Move the Markdown text into developer_instructions by hand. For all projects use ~/.codex/agents/
Manually: put the files in place
.cursor/agents/agent_read.mdCursor also reads .claude/agents/
Step by step, with commands for each platform: Claude Code subagents: how to create one, call it and keep context clean
Questions about agent_read
How do I install agent_read in Claude Code?
Add the subagent file at .claude/agents/agent_read.md.
What does agent_read do?
It reads a selected agent’s recent conversation, including user messages, agent replies, and short records of tool calls. Another agent can use that history to answer follow-up questions without starting a new turn.
What information does agent_read need?
It needs machine and session_id. The session ID can be full or its first 8 characters; turns controls the number of recent exchanges, with a default of 2 and a maximum of 10.
Source text
agent_read.md · MIT
Read the recent conversation of one agent (from agents_list): the user's messages, what the agent said, tool calls in short form. `turns` = how many recent exchanges (default 2, max 10). Use it to answer the user's follow-ups about work an agent or task already did (its conclusion, a PR link, why it stopped) instead of asking the agent with agent_send, which makes it take another turn while the user waits (and a stopped agent takes no messages: it is read-only).
```json
{
"type": "object",
"required": ["machine", "session_id"],
"properties": {
"machine": {
"type": "string"
},
"session_id": {
"type": "string",
"description": "full id or its first 8 characters"
},
"turns": {
"type": "integer"
}
}
}
```
© nwverse · MIT · nwverse/persona