Tencent open-sources LoopForge: turning one-off coding-agent runs into team workflows with approval gates
Synopsis
Tencent has open-sourced LoopForge, a self-hosted workspace where a team describes a requirement to Main Bot in chat and it opens a Docker sandbox, checks out repositories, delegates to project roles and opens the PR. Previously, asking a coding agent to fix code left the result dependent on that run, which might skip the plan, forget tests or review its own code. LoopForge saves a proven way of working as an SOP, a standard workflow of agents, scripts and human gates, so later requirements follow the same stages and checks, with approvals, node status and tool traces kept in one place.
Interpretation
Tencent released LoopForge, a self-hosted coding-agent workspace where a team describes a requirement in chat and agents carry it from repository checkout to pull request. Asking a coding agent directly to change code leaves the result dependent on that run, which might skip the plan, forget the tests, review its own code or stop halfway unnoticed. LoopForge tracks both paths for the same requirement, so every change, whether made in chat or by an SOP, links back to what was asked for and counts toward delivery and cost reports.
The workspace splits a task into planning, implementation, review and testing roles, each with its own prompt, executor, model, rules, Skills and MCP tools. A role can run as an SOP node or as a subagent in chat, separating steps that would otherwise be mixed inside one run. Executors cover ten coding-agent CLIs, including CodeBuddy, Codex, Claude Code, OpenCode, Cursor CLI, Antigravity, Kimi Code, Pi, DeepSeek Harness and ZCode, with Docker sandboxes or the team's own machine as the runtime.
A way of working that proves itself can be saved as an SOP, a standard workflow of agents, scripts and human gates, so the next requirement is delivered the same way. Built-in backend, frontend, bug-fix and product SOPs give a starting point, and people can read the report, approve, or send work back to a chosen step at approval gates. Node status, tool traces, logs and artifacts sit in one place, so a run can be paused, a failed step retried, or work rerun from any node; the system is deployed on the team's own infrastructure under the MIT license.
Perspective
The workspace targets teams that ship software on their own servers and want agent steps fixed as repeatable workflows; installation requires Git, Docker, Compose, Python 3 and curl, and package and image downloads need internet access.
The public description gives no quantitative comparison of SOPs against one-off chat runs in delivery quality or cost, and reports no results from use on real team projects.