️ Tool
===================

Opening: Crit is a local-first feedback and review utility focused on human-in-the-loop iteration with AI coding agents. It provides a PR-style review surface for agent plans and code, keeps previous comments visible across rounds, and notifies agents when reviews finish so they can iterate automatically.

Key Features:
• Inline Comments: PR-style inline comments on selected line ranges, preserved across rounds.
• Diff Views: Side-by-side git diffs and round diffs for tracking changes between agent iterations.
• Agent Integrations: Integrations described for Claude Code, GitHub Copilot, Aider, Cline and Windsurf; designed to operate with any agent that reads files.
• Rich Rendering: Syntax highlighting across 190+ languages and interactive Mermaid diagram rendering for plans/specs.
• Editor Shortcuts: Vim-style keybindings and keyboard-driven review flow, plus sharing of review links.

Technical Implementation:
• Local-first Architecture: The tool emphasizes local operation and file-based listening rather than cloud-only workflows; it monitors repository files and skill files so agents can pick up feedback without centralized authentication.
• Feedback Loop Mechanism: Reviews generate structured comments that the connected agent can consume; the user finalizes a review (Finish Review) and the agent is signalled to edit files and produce a new round. The tool stores round metadata to compute diffs between revisions.
• Pluggable Agent Hooks: Integration points are implemented as skill files or agent-facing rules so multiple agent implementations can be supported without changing the core review UI.

Use Cases:
• Team members reviewing proposed agent plans or code before the agent performs changes.
• Iterative spec-to-code workflows where visible comments must persist across agent edits.
• Maintaining human oversight over multi-round agent workflows in private/local environments.

Limitations and Considerations:
• The tool is oriented toward local workflows and file-based agents; workflows that require centralized cloud coordination or hosted UIs may require additional integration.
• Features that mention sessions or tokens (e.g., JWT tokens in example comments) indicate practices to review for production-grade session management but do not prescribe specific authentication patterns.
• The repository showcases integrations but specific agent behavior depends on each agent's ability to read feedback files and react programmatically.

tool #ai #feedback #localfirst

Source: https://crit.md/