varpet

Autonomous bug-fixing agent. Reads Jira tickets, analyzes repositories, applies minimal patches, and creates pull requests.

Installation

npm install -g varpet

Quick Start

# Fix a bug from a Jira ticket
varpet fix --jira PROJ-123 --repo .

# Fix with a specific base branch
varpet fix --jira PROJ-123 --repo . --base develop

# Dry run (simulate without making changes)
varpet fix --jira PROJ-123 --repo . --dry-run

# Resume a previous run
varpet resume --run-id <uuid>

# Check status of a run
varpet status --run-id <uuid>

Configuration

Varpet requires environment variables for connecting to external services. Create a .env file in your project root or export them in your shell:

Required

# Jira (Cloud: use email + API token, Server: use PAT)
JIRA_URL=https://your-org.atlassian.net
[email protected]
JIRA_API_TOKEN=your-jira-api-token

# GitHub
GITHUB_TOKEN=your-github-token

# At least one LLM provider key
ANTHROPIC_API_KEY=your-key
# or
OPENAI_API_KEY=your-key
# or
GEMINI_API_KEY=your-key

Optional

# Model configuration (defaults shown)
VARPET_REASONING_MODEL=anthropic/claude-sonnet-4-20250514
VARPET_PATCHING_MODEL=anthropic/claude-sonnet-4-20250514
VARPET_MAX_ITERATIONS=5
VARPET_CONFIDENCE_THRESHOLD=0.7
VARPET_LOG_LEVEL=INFO

Commands

varpet fix

Fix a bug described in a Jira ticket.

| Option | Short | Default | Description | |----------------------|-------|---------|------------------------------------------| | --jira | -j | required | Jira issue key (e.g., PROJ-123) | | --repo | -r | . | Path to repository (or URL to clone) | | --base | -b | main | Base branch to create fix branch from | | --dry-run | | false | Simulate without modifying files | | --max-iterations | -n | 5 | Max fix/validate iterations | | --model | -m | | Override reasoning model | | --verbose | | false | Enable debug logging |

varpet resume

Resume a previous run from its saved state.

| Option | Short | Default | Description | |-------------|-------|---------|------------------------| | --run-id | | required | Run ID to resume | | --repo | -r | . | Path to repository | | --verbose | | false | Enable debug logging |

varpet status

Show the status of a previous run.

| Option | Short | Default | Description | |------------|-------|---------|------------------------| | --run-id | | required | Run ID to check | | --repo | -r | . | Path to repository |

How It Works

1. Fetch Jira - Reads the bug ticket and extracts signals (stack traces, error messages)
2. Prepare Repo - Clones/verifies the repository and creates a work branch
3. Build Index - Builds an AST + embedding index of the codebase
4. Retrieve Context - Finds relevant code using ripgrep + semantic search
5. Investigate - LLM analyzes root cause, generates hypotheses
6. Plan Patch - LLM creates a minimal patch plan
7. Apply Patch - Applies changes with fallback strategies
8. Generate Tests - LLM writes a regression test
9. Validate - Runs tests, computes confidence score
10. Create PR - Commits, pushes, creates GitHub PR (or retries on failure)

Supported Platforms

| Platform | Architecture | |----------------------|-------------| | macOS | ARM64 (Apple Silicon) | | macOS | x64 (Intel) | | Linux | x64 | | Windows | x64 |

License

Apache-2.0