← All Skills
📝
Plannotator Visual Agent Review Skills
Open agent plans, specs, Markdown, HTML, local diffs, and remote pull requests in a browser-based review surface, annotate or approve them, and return structured feedback directly to Claude Code, Codex, and other agents.
Key Features
- Visual plan and document annotation
- Local diff, GitHub PR, and GitLab MR review
- Strict approval gates and JSON outcomes
- Local history with optional encrypted sharing
Installation
On macOS, Linux, or WSL, run curl -fsSL https://plannotator.ai/install.sh | bash. The installer downloads the binary from GitHub Releases and configures hooks, skills, and commands for detected agents. Claude Code can also install plannotator@plannotator from the backnotprop/plannotator marketplace; Codex setup enables its experimental Stop hook.
How to Use
In Codex, use $plannotator-review, $plannotator-annotate, and $plannotator-last, or run !plannotator review. For a machine-checkable file approval, use plannotator annotate <file> --gate --json --require-approval. Do not treat closing a plain annotate session as approval.
Requirements
Requires the plannotator binary and a local browser. Plans, diffs, annotations, and history stay local by default with no telemetry, but app surfaces check GitHub for releases. URL annotation uses Jina Reader by default, remote reviews use authenticated CLIs, Ask AI sends selected context to the configured provider, and sharing can upload content or ciphertext. Disable those features with PLANNOTATOR_AI=disabled and PLANNOTATOR_SHARE=disabled. Codex Stop hooks remain experimental.
Related Skills