Keep the Human in the Loop of coding
See the code
Review AI agent code changes before commit.
GitHub revolutionized how humans collaborate on code.
GitHuman defines how humans review code written by AI.
AI coding agents write code. But the traditional PR workflow assumes humans are the authors. By the time AI-generated code reaches a pull request, you've already committed to the approach. Review happens too late.
GitHuman moves the review checkpoint to where it belongs: the staging area.
Before git commit, you get a proper review interface—not a wall of terminal diff output. Add comments, track issues, and make informed decisions about what the AI produced.


githuman ask to ask a human for review, wait, and hand their feedback back to the agentnpm install -g githuman
Or run directly:
npx githuman@latest serve
# Stage your changes (from AI agent or manual edits)
git add .
# Start the review interface
githuman serve
This opens a web interface at http://localhost:3847 where you can review your staged changes before committing.
githuman ask "Please review the parser refactor"
githuman ask starts or reuses GitHuman, opens a dedicated ask page, waits for the human to finish reviewing, and then prints the new todos, comments, and review status updates for the agent.
The handoff is completed in the UI with a dedicated Continue assistant button.
GitHuman provides an agent skill that teaches AI coding agents when and how to use GitHuman for reviewing changes.
Install it with the skills CLI:
npx skills add mcollina/githuman-skills
githuman serve [options]
Options:
-p, --port <port> Port to listen on (default: 3847)
--host <host> Host to bind to (default: localhost)
--open Auto-open browser
--no-open Don't open browser automatically
--https Enable HTTPS
--no-https Disable HTTPS
--cert <path> Use a custom TLS certificate
--key <path> Use a custom TLS private key
--auth [token] Enable auth, optionally with a provided token
-v, --verbose Enable verbose logging
-h, --help Show help
githuman ask [message] [options]
Options:
-p, --port <port> Port to use for GitHuman
--host <host> Host to bind to when starting GitHuman
--open Open the browser automatically
--no-open Don't open the browser automatically
--review <id> Scope feedback to a specific review
--interval <ms> Polling interval in milliseconds
--json Output final feedback as JSON
-h, --help Show help
githuman list [options]
Options:
--status <status> Filter by status (in_progress|approved|changes_requested)
--json Output as JSON
-h, --help Show help
githuman export <review-id|last> [options]
Arguments:
review-id The ID of the review, or "last" for most recent
Options:
-o, --output <file> Output file path (default: stdout)
--no-resolved Exclude resolved comments
--no-snippets Exclude diff snippets
-h, --help Show help
githuman todo <subcommand> [options]
Subcommands:
add <content> Add a new todo item
list List all todos (pending by default)
done <id> Mark todo as completed
undone <id> Mark todo as not completed
remove <id> Delete a todo
clear --done Remove all completed todos
Options:
--review <id> Scope todo to a specific review
--all Show all todos (not just pending)
--done Filter to show only completed todos
--json Output as JSON
-h, --help Show help
githuman serve - Opens the review interfacegithuman askContinue assistant when feedback is readygithuman ask exits with the new feedback so the agent can continuegit addgithuman serve+ button that appears| Key | Action |
|---|---|
j | Next file |
k | Previous file |
Esc | Cancel / Close |
The server exposes a REST API with OpenAPI documentation at /docs.
Set a token to require authentication:
# Via CLI flag
githuman serve --auth mysecrettoken-that-is-at-least-32-characters
# Auto-generate a token
githuman serve --auth
# Via environment variable
GITHUMAN_TOKEN=mysecrettoken-that-is-at-least-32-characters githuman serve
Clients must include the token in the Authorization header:
Authorization: Bearer mysecrettoken-that-is-at-least-32-characters
GitHuman can read repo-local defaults from .githuman/config.json.
{
"host": "localhost",
"port": 3847,
"https": false,
"open": true
}
Config precedence is:
.githuman/config.jsonGitHuman stores local data in:
<repository>/.githuman/
Important files include:
reviews.db - reviews, comments, and todosconfig.json - optional repo-local CLI defaultsThis directory is typically gitignored.
# Clone the repository
git clone https://github.com/mcollina/local-code-reviewer.git
cd local-code-reviewer
# Install dependencies
npm install
# Run server in watch mode
npm run dev:server
# Run web dev server (Vite)
npm run dev
# Run all tests
npm test
In the age of AI coding assistants, someone needs to review the code before it's committed. That someone is you. GitHuman is the human checkpoint in an AI-assisted workflow.
MIT License for the code.
The GitHuman logo is Copyright (c) Matteo Collina, All Rights Reserved.
2,103 followers · starred Jan 2026
2,680 followers · starred Jan 2026
226 followers · starred Apr 2026
95 followers · starred Jan 2026
Keep the Human in the Loop of coding
See the code
Review AI agent code changes before commit.
GitHub revolutionized how humans collaborate on code.
GitHuman defines how humans review code written by AI.
AI coding agents write code. But the traditional PR workflow assumes humans are the authors. By the time AI-generated code reaches a pull request, you've already committed to the approach. Review happens too late.
GitHuman moves the review checkpoint to where it belongs: the staging area.
Before git commit, you get a proper review interface—not a wall of terminal diff output. Add comments, track issues, and make informed decisions about what the AI produced.


githuman ask to ask a human for review, wait, and hand their feedback back to the agentnpm install -g githuman
Or run directly:
npx githuman@latest serve
# Stage your changes (from AI agent or manual edits)
git add .
# Start the review interface
githuman serve
This opens a web interface at http://localhost:3847 where you can review your staged changes before committing.
githuman ask "Please review the parser refactor"
githuman ask starts or reuses GitHuman, opens a dedicated ask page, waits for the human to finish reviewing, and then prints the new todos, comments, and review status updates for the agent.
The handoff is completed in the UI with a dedicated Continue assistant button.
GitHuman provides an agent skill that teaches AI coding agents when and how to use GitHuman for reviewing changes.
Install it with the skills CLI:
npx skills add mcollina/githuman-skills
githuman serve [options]
Options:
-p, --port <port> Port to listen on (default: 3847)
--host <host> Host to bind to (default: localhost)
--open Auto-open browser
--no-open Don't open browser automatically
--https Enable HTTPS
--no-https Disable HTTPS
--cert <path> Use a custom TLS certificate
--key <path> Use a custom TLS private key
--auth [token] Enable auth, optionally with a provided token
-v, --verbose Enable verbose logging
-h, --help Show help
githuman ask [message] [options]
Options:
-p, --port <port> Port to use for GitHuman
--host <host> Host to bind to when starting GitHuman
--open Open the browser automatically
--no-open Don't open the browser automatically
--review <id> Scope feedback to a specific review
--interval <ms> Polling interval in milliseconds
--json Output final feedback as JSON
-h, --help Show help
githuman list [options]
Options:
--status <status> Filter by status (in_progress|approved|changes_requested)
--json Output as JSON
-h, --help Show help
githuman export <review-id|last> [options]
Arguments:
review-id The ID of the review, or "last" for most recent
Options:
-o, --output <file> Output file path (default: stdout)
--no-resolved Exclude resolved comments
--no-snippets Exclude diff snippets
-h, --help Show help
githuman todo <subcommand> [options]
Subcommands:
add <content> Add a new todo item
list List all todos (pending by default)
done <id> Mark todo as completed
undone <id> Mark todo as not completed
remove <id> Delete a todo
clear --done Remove all completed todos
Options:
--review <id> Scope todo to a specific review
--all Show all todos (not just pending)
--done Filter to show only completed todos
--json Output as JSON
-h, --help Show help
githuman serve - Opens the review interfacegithuman askContinue assistant when feedback is readygithuman ask exits with the new feedback so the agent can continuegit addgithuman serve+ button that appears| Key | Action |
|---|---|
j | Next file |
k | Previous file |
Esc | Cancel / Close |
The server exposes a REST API with OpenAPI documentation at /docs.
Set a token to require authentication:
# Via CLI flag
githuman serve --auth mysecrettoken-that-is-at-least-32-characters
# Auto-generate a token
githuman serve --auth
# Via environment variable
GITHUMAN_TOKEN=mysecrettoken-that-is-at-least-32-characters githuman serve
Clients must include the token in the Authorization header:
Authorization: Bearer mysecrettoken-that-is-at-least-32-characters
GitHuman can read repo-local defaults from .githuman/config.json.
{
"host": "localhost",
"port": 3847,
"https": false,
"open": true
}
Config precedence is:
.githuman/config.jsonGitHuman stores local data in:
<repository>/.githuman/
Important files include:
reviews.db - reviews, comments, and todosconfig.json - optional repo-local CLI defaultsThis directory is typically gitignored.
# Clone the repository
git clone https://github.com/mcollina/local-code-reviewer.git
cd local-code-reviewer
# Install dependencies
npm install
# Run server in watch mode
npm run dev:server
# Run web dev server (Vite)
npm run dev
# Run all tests
npm test
In the age of AI coding assistants, someone needs to review the code before it's committed. That someone is you. GitHuman is the human checkpoint in an AI-assisted workflow.
MIT License for the code.
The GitHuman logo is Copyright (c) Matteo Collina, All Rights Reserved.
2,103 followers · starred Jan 2026
2,680 followers · starred Jan 2026
226 followers · starred Apr 2026
95 followers · starred Jan 2026