Crit: Local-first, single-binary CLI for reviewing agent plans and diffs
Crit is a local-first review tool packaged as a single binary. It opens your file or code diff in a browser with a GitHub-inspired interface, lets you leave comments, then your agent acts on the feedback automatically. Repeat until satisfied — all before hitting GitHub.
Core loop: comment, agent updates, diff refreshes
Drag across a range of lines, leave comments, hit Finish Review. The agent picks up the feedback, edits the file, and Crit reloads with the updated diff. Previous comments stay visible; resolve only when you're satisfied. Works for both plans and code — same UI, beautifully rendered markdown with syntax highlighting for 190+ languages.
Install in one command
$ brew install crit $ go install github.com/tomasz-tomczyk/crit@latest $ nix profile install github:tomasz-tomczyk/crit # Windows: PS> iwr https://github.com/tomasz-tomczyk/crit/releases/latest/download/crit-windows-amd64.exe -OutFile crit.exe
Then run crit or crit plan.md. Pre-built binaries for ARM64, WSL also available.
Integrates with 11 agents
One-command setup for Claude Code, Cursor, Copilot, Codex, OpenCode, Gemini, Qwen, Hermes, Windsurf, Cline, Aider. Example:
$ claude plugin add tomasz-tomczyk/crit $ claude plugin install crit@crit $ crit install cursor $ crit install github-copilot $ crit install codex
See full setup docs for each agent.
Multi-round and concurrent reviews
Iterate over multiple rounds per file. Run several reviews in parallel without stepping on each other — useful when juggling multiple agents.
Optional sharing and self-hosting
Reviews stay local by default. One click generates a public link — no install, no login for viewers. You can self-host the share target (crit-web) for privacy. Without sharing, reviews auto-delete after 30 days of inactivity.
Small quality-of-life features
- Vim keybindings
- Git diffs (split & unified)
- Round-over-round diffs
- Mermaid diagrams
- Table of contents
- Dark & light themes
- Draft autosave
- Insert suggestions
Who it's for
Developers who frequently review AI-generated plans and code diffs and want a dedicated local UI to batch feedback and iterate without leaving their workflow.
📖 Read the full source: HN AI Agents
👀 See Also
Open-source trust scoring hook for Claude Code monitors sessions, blocks protected paths
A developer built a Python hook that scores every Claude Code session on reliability, scope, and cost dimensions, blocks access to protected paths like .env files, and hash-chains events for tamper detection. The single-file tool is available on GitHub.
Qwen 3.6 27B Quantization Benchmark: Q4_K_M Beats Q8_0 on Practical Tradeoffs
Evaluated Qwen 3.6 27B across BF16, Q4_K_M, and Q8_0 GGUF quants on HumanEval, HellaSwag, and BFCL. Q4_K_M delivers near-BF16 scores with 48% less RAM, 1.45x speed, and 68.8% smaller file size.
Open Source Grafana Dashboard Tracks Claude Code Costs and Usage via OpenTelemetry
An SRE built a free Grafana dashboard to visualize Claude Code spend, token usage, cache hit ratios, and edit decisions by pulling OpenTelemetry metrics into Prometheus-compatible backends.
SpecLock: MCP Server for Enforcing AI Coding Constraints
SpecLock is an open-source MCP server that remembers project constraints across sessions and blocks AI coding agents from violating them. Claude independently tested it with 100 adversarial tests, scoring 100/100 with zero false positives and 15.7ms per check.