CodexAgent

Codex Agent

langbot-teamCodexAgent
Plugin22 downloadsAgent Runner1

Run Codex CLI as a LangBot AgentRunner.

Codex Agent

Overview

Run Codex CLI as a LangBot AgentRunner.

Package information

Capabilities

  • Enabled: streaming, tool calling, knowledge retrieval, steering
  • Not declared: multimodal input, interrupt

Configuration

FieldTypeRequiredDefault
daemon-enabledbooleanNofalse
daemon-hoststringNo127.0.0.1
daemon-portintegerNo8768
daemon-tokensecretNoEmpty
locationselectYeslocal
workspacestringNoEmpty
commandstringNocodex
args-jsonstringNo[]
env-jsonstringNo{}
ssh-targetstringNoEmpty
ssh-portintegerNo22
daemon-idstringNoEmpty
timeoutintegerNo1800
streamingbooleanNotrue
reuse-sessionbooleanNotrue
approval-policyselectNonever
sandbox-modeselectNodanger-full-access
knowledge-basesknowledge-base-multi-selectorNo[]
langbot-assets-enabledbooleanNotrue
mcp-bridge-transportselectNoauto
mcp-servers-jsonstringNo[]

Host permissions

  • tools: detail, call
  • knowledge_bases: retrieve
  • history: page

Installation and usage

  1. Install the plugin from the LangBot plugin marketplace.
  2. Select the Runner ID below in the Pipeline AgentRunner selector.
  3. Fill in connection settings from the table and store credentials in secret fields in the admin UI.

Security and limitations

  • The default configuration starts Codex with approvalPolicy=never and sandbox=danger-full-access and does not wait for interactive approval. Use the default only with trusted workspaces and a constrained operating-system account.
  • The runner can use only LangBot resources authorized for the current run.
  • Availability, model abilities, and rate limits depend on the external service.
  • See the full Chinese README at the package root for advanced behavior and product-specific limitations.

Structured interactions

The runner registers a real Codex app-server dynamic tool named ask_user_question and also accepts Codex's native item/tool/requestUserInput request. Both are converted to LangBot's provider-neutral interaction.requested contract, so Lark, DingTalk, and other delivery adapters can render the same fields and actions.

The app-server process is stopped while the user is answering. The submission starts a new AgentRun and resumes the same Codex thread. Codex app-server does not support replying to an old JSON-RPC tool request from a new process, so the answer is delivered as the next authoritative user turn rather than injected as the old tool result. This is a provider transport limitation; the interaction is still a real model tool call and no process is kept waiting.

Codex app-server command-execution and file-change approval requests are also converted to action-only confirmation cards. The previous turn is cancelled while the user decides. Approval resumes the same thread and grants exactly one matching retry; a different command or file change requires a new approval.

Use Codex Agent in Feishu/Lark bots, DingTalk bots, WeCom bots, WeChat bots, QQ bots, Slack bots, Discord bots, Telegram bots, LINE Bots, and Matrix Bots.

Host your bot on LangBot Cloud and connect this extension to the collaboration platforms your team already uses.

Open LangBot Cloud

Comments