codecheck

A web application that reviews GitHub repositories using Claude Code. Paste a repo URL, pick a review prompt, and get streaming code analysis with actionable recommendations.

Quick Start

pip install -r requirements.txt
cp .env.default .env        # edit as needed
python app.py
# Open http://localhost:8000

How It Works

1. Enter a GitHub repository URL (e.g. user/repo or https://github.com/user/repo)
2. Select a prompt template from the dropdown, or write your own
3. Click Evaluate Now — the repo is cloned and analyzed by Claude Code in real time
4. Watch the live terminal chatter (tool calls, file reads, commands) as Claude Code works
5. When done, the final report appears below with rendered markdown and syntax highlighting
6. Ask follow-up questions — Claude Code continues in the same session with full context
7. Optionally file the results as a GitHub issue (requires gh CLI authenticated)

Features

• Live streaming — SSE-based real-time output with a terminal-like chatter box showing Claude Code's tool calls
• Follow-up questions — continue the conversation in the same Claude Code session (--continue)
• Persistent history — past evaluations stored in the browser for 30 days, deduplicated per repo
• Generated files — markdown files Claude Code creates during analysis are linked as chips, persisted under /tmp for 30 days
• Copy markdown — one-click copy of the full report or individual generated files
• File as GitHub Issue — post results directly to the analyzed repo's issue tracker
• Dark UI — minimal single-page app with OSU Beaver Orange accent

Claude Code Backends

The app requires both the Claude Code CLI and an authentication method. If the CLI is not installed or no credentials are configured, the app will not function. Initial evaluations use Sonnet; follow-up questions use Opus.

The app checks for the claude binary in ~/.local/bin/claude, ~/bin/claude, then PATH. When found, it runs Claude Code via subprocess with --output-format stream-json --verbose — this gives the richest experience with tool calls, file operations, and the full agent loop streamed live.

You must configure one of the following authentication methods:

Option 1: Anthropic API Key

The simplest setup — just set your API key:

ANTHROPIC_API_KEY=sk-ant-...

Option 2: AWS Bedrock

Uses the Anthropic SDK via AWS Bedrock for authentication.

First configure your AWS credentials:

aws --profile codecheck configure
# Enter your AWS Access Key ID, Secret Access Key, and region (us-west-2) when prompted

Then set in .env:

CLAUDE_CODE_USE_BEDROCK=1
AWS_PROFILE=codecheck
AWS_DEFAULT_REGION=us-west-2
ANTHROPIC_DEFAULT_OPUS_MODEL=global.anthropic.claude-opus-4-7
ANTHROPIC_DEFAULT_SONNET_MODEL=global.anthropic.claude-sonnet-4-6

Option 3: Azure AI Foundry

Uses the Anthropic SDK via Microsoft Azure AI Foundry for authentication.

CLAUDE_CODE_USE_FOUNDRY=1
ANTHROPIC_FOUNDRY_BASE_URL=https://<resource>.services.ai.azure.com
ANTHROPIC_FOUNDRY_API_KEY=your-azure-api-key
ANTHROPIC_DEFAULT_OPUS_MODEL=claude-opus-4-7
ANTHROPIC_DEFAULT_SONNET_MODEL=claude-sonnet-4-6

See Anthropic on Azure docs.

SDK Fallback (no CLI)

If the Claude Code CLI is not installed, the app falls back to using the Anthropic SDK directly (requires Option 2 or 3 above plus pip install anthropic[bedrock] or pip install anthropic). This path builds a text context from the repo's files and streams the response, but doesn't have Claude Code's tool-use capabilities.

Prompt Templates

Templates are .prmpt files where the first line is the display name and the rest is the prompt body. They are loaded from two locations (merged, user-local wins on name conflicts):

• prompts/ — shipped defaults (code quality, multi-GPU, security)
• ~/.codecheck/prompts/ — your own custom templates

Configuration

Copy .env.default to .env and adjust:

cp .env.default .env

All settings with defaults are documented in .env.default. Key variables:

Variable	Default	Description
PORT	8000	Server port
CLAUDECODE	—	Set to skip CLI (auto-set inside Claude Code sessions)
ANTHROPIC_API_KEY	—	Anthropic API key (Option 1)
CLAUDE_CODE_USE_BEDROCK	—	Set to 1 to use AWS Bedrock (Option 2)
AWS_PROFILE	codecheck	AWS profile (set up with aws --profile codecheck configure)
AWS_DEFAULT_REGION	us-west-2	AWS region for Bedrock
CLAUDE_CODE_USE_FOUNDRY	—	Set to 1 to use Azure AI Foundry (Option 3)
ANTHROPIC_FOUNDRY_BASE_URL	—	Azure AI Foundry endpoint URL
ANTHROPIC_FOUNDRY_API_KEY	—	Azure AI Foundry API key
ANTHROPIC_DEFAULT_OPUS_MODEL	per-backend	Opus model for follow-ups (SDK fallback only)
ANTHROPIC_DEFAULT_SONNET_MODEL	per-backend	Sonnet model for initial eval (SDK fallback only)
GH_TOKEN / GITHUB_TOKEN	—	GitHub token for higher clone rate limits

Deployment

The app is designed for appmotel (systemd + Traefik PaaS):

sudo -u appmotel appmo add codecheck dirkpetersen/codecheck main
sudo -u appmotel appmo env codecheck   # set env vars
sudo -u appmotel appmo restart codecheck

The PORT env var is set automatically by appmotel. Uvicorn auto-reload is disabled when running under systemd (SYSTEMD_EXEC_PID is set).

Tech Stack

• Python 3.12+ / FastAPI with async SSE streaming
• Claude Code CLI (claude --output-format stream-json --verbose) or Anthropic SDK (Bedrock / Azure)
• marked.js + highlight.js for client-side markdown rendering
• Dark minimal single-page UI with localStorage-based history

License

MIT