update
May 14, 2026
By Teun
PlanBridge lets developers review coding agent plans locally
PlanBridge is an open-source tool that intercepts coding agents’ plans before code is written, letting users annotate and request changes line by line. It runs locally, integrates with Claude Code and Codex CLI, and keeps plan content on the user’s machine.
PlanBridge is an open-source tool for reviewing AI coding agent plans before they become code. According to the project’s Show HN post, it intercepts an agent’s plan, opens it in the browser, and lets users mark up individual lines with precise feedback before the first line of code is written.
The tool is built by ContextBridge and runs locally. The post says it hooks into Claude Code and Codex CLI, and can also be used with other agents that run shell commands and parse JSON. In those cases, the agent can pipe a plan to contextbridge plan and act on the response.
The installer is designed to be simple. The project provides a one-line install script, a Homebrew option, and a tarball fallback into $HOME/.local/bin. After installation, contextbridge install detects supported AI coding harnesses and adds the review hooks automatically.
The workflow is meant to fit into a normal coding session. First, the user installs PlanBridge once. Then they keep using their coding harness as usual, and when the agent produces a plan, the browser opens for review. The user can annotate the plan, approve it, or request changes, and that decision is passed back to the harness.
The project lists explicit hook points for supported tools. For Claude Code, it uses PermissionRequest:ExitPlanMode. For Codex CLI, it uses Stop. That means the review step happens before execution, rather than after an agent has already started writing files or running commands.
Privacy is a major part of the pitch. The post says plan content stays on the user’s machine and is not sent to a remote backend, account, or API keys. It also says the CLI sends anonymous product analytics and crash reports by default unless the user opts out with DO_NOT_TRACK=1 or CONTEXTBRIDGE_TELEMETRY_DISABLED=1.
ContextBridge says the tool is open source and invites users to inspect the quickstart, install guide, usage docs, privacy notes, Slack community, and GitHub issues. The post also says PRs and issues are welcome, and that local setup, the development loop, and the release process are documented in CONTRIBUTING.md.
For teams using AI coding agents, the main change is that plan review becomes a first-class step instead of an informal glance. PlanBridge is trying to make AI-assisted coding closer to code review, but for the agent’s intent before implementation starts.
