Docs / Coding agents
Coding agents and your own agents
| Agent | Guide | Verified in the real product |
|---|---|---|
| Claude Code | claude-code.md | ✅ Claude Code 2.1 (MCP flow and hook) |
| OpenAI Codex CLI | codex.md | ✅ codex-cli 0.157 (MCP flow, and hook blocking with self-correction) |
| OpenCode | opencode.md | ✅ OpenCode 1.18 (MCP flow and plugin) |
| Gemini CLI | gemini-cli.md | ✅ Gemini CLI 0.64 (MCP flow and hook) |
| Cursor | cursor.md | ⚠️ hook tested with sample payloads only |
| Claude Desktop, VS Code, Windsurf, Zed, other MCP clients | mcp-clients.md | ⚠️ documented configs, not tested end to end |
| LangGraph, OpenAI Agents SDK, Claude Agent SDK, Vercel AI SDK | sdks.md | ✅ SDKs tested against a live executor |
Every agent gets the same model:
- The agent proposes. Changeward plans with an exact dry run, and policy decides.
- Changeward applies with an undo path.
- A hook or plugin stops the agent from going around Changeward with raw
psql,kubectlorterraform.
Edit this page on GitHub · synced from 89c04c9