Recon Magic

Recon Magic provides AI-powered automations for smart contract fuzzing. While Fuzzing Jobs let you run fuzzing tools in the cloud with extensive features and functionality, Recon Magic offers AI agents that automate the setup and optimization process.

Access Recon Magic at: getrecon.xyz/dashboard/magic

Works with both public and private repositories. For private repos, install the Recon GitHub App first.

Running Magic Jobs

This section covers how Magic Jobs work, regardless of which specific workflow you choose.

How It Works

When you start a Magic Job, the following happens:

Repository Creation: A new GitHub repository is created for your job results (e.g., your-repo-processed-1734567890)
User Invitation: You (and any other GitHub users you specify) are automatically invited as collaborators to the new repository
Initial Push: Your original code is pushed to the new repository as the initial commit
Workflow Execution: The AI workflow runs, making changes and committing them step by step
Results: View the diff, generated documentation, and any artifacts directly in the dashboard

Repository Access

Automatic Invitations

When you start a job, you are automatically invited to the results repository. The invitation is sent before the workflow begins, so you can follow along in real-time.

Inviting Additional Users

You can invite other GitHub users to the results repository at any time:

Navigate to your job's detail page in the dashboard
Find the "Invite GitHub user to repository" form
Enter the GitHub username of the person you want to invite
Click "Invite"

The invited user will receive a GitHub collaboration invite with push access to the repository.

Code Updates and Commits

Magic Jobs commit changes after each workflow step. This provides:

Transparency: See exactly what the AI modified at each step
Traceability: Each step has its own commit hash you can reference
Real-time progress: Changes are pushed to GitHub as they happen

In the job detail page, you can:

View the workflow progress with step-by-step completion status
Click on commit hashes to view them directly on GitHub
Expand each step to see which files were changed
Click on file names to view them at that specific commit

Stopping a Job Early

If you need to stop a running job:

Navigate to the job's detail page
Click the "Stop Job" button (visible while the job is running)
The job will gracefully stop after the current step completes

The "Stop requested" badge will appear once your request is received. The job will finish its current step, commit any changes, and then stop. You'll still have access to all the work completed up to that point.

Job Statuses

Queued: Job is waiting to be processed
WIP (Work in Progress): Job is currently running
Done: Job completed successfully
Stopped: Job was stopped early by user request
Error: Job encountered an error during execution

See below for workflow-specific tutorials and details on each Magic Job type.

V2 Workflows

The V2 AI agents are designed to be used in sequence. You can start from code as long as it's somewhat valid. Each step builds on the previous one:

Scout V2 (`workflow-fuzzing-scouting`)

Give Scout V2 any repository and it will essentially use the Recon Extension for you. It analyzes your codebase and identifies invariants in your smart contracts.

Input: Any GitHub repository

Setup V2 (`workflow-fuzzing-setup`)

Setup V2 fixes compilation issues and creates a professional fuzzing setup so your fuzzing campaign will be useful.

Input: Repository (can have compilation issues) Output: Fixed compilation + professional fuzzing setup

Coverage V2 (`workflow-fuzzing-coverage`)

Coverage V2 requires valid compilation and a valid setup. It then works to push your code toward better coverage.

Input: Repository with valid compilation and valid setup Output: Improved coverage for your fuzzing campaign

How it works: Coverage V2 runs a multi-step workflow that:

Extracts target functions from your fuzzing suite
Analyzes build artifacts to identify all functions that need coverage
Creates clamped handlers with meaningful values for better fuzzing
Creates shortcut handlers to help the fuzzer reach coverage faster
Runs fuzzing and evaluates coverage
Iteratively improves handlers until coverage targets are met

Each step automatically commits changes to git, so you can track exactly what the AI modified.

Using the Workflows

Start with Scout V2 to analyze your repository and identify invariants
Run Setup V2 to fix any compilation issues and establish a solid fuzzing foundation
Use Coverage V2 to maximize your test coverage

Automatic chaining of these workflows is coming soon. Currently, running them separately provides better reliability.

Recon Book