remy-agent 0.1.0

AI-Powered Codebase Bug & Vulnerability Scanning CLI

Remy Agent

AI-Powered Codebase Bug & Vulnerability Scanning CLI

---

What is Remy?

Remy is a terminal-native AI security and bug-scanning agent. It performs static application security testing (SAST) plus an optional AI-assisted logic/bug detection pass across your codebase, then generates a single, structured Fix Prompt — ready to paste into any AI coding agent of your choice.

Remy is the finder and prompt-compiler, not the editor. That makes it lightweight, model-agnostic, and safe to run on any codebase without needing write access.

---

Features

• Multi-engine SAST — Python AST rules for injection, insecure deserialization, weak crypto, and more
• Secrets scanner — regex + Shannon entropy detection for API keys, tokens, and credentials across all file types
• API surface auditor — finds unauthenticated routes, missing rate limiting, and overly permissive CORS
• Auth/bypass scanner — detects JWT bypass, IDOR patterns, non-constant-time comparisons, client-controlled admin flags
• Dependency scanner — cross-references your packages against the OSV vulnerability database
• LLM logic-bug pass (--deep) — sends code to your AI provider to catch race conditions, edge cases, and business logic flaws
• Multi-provider AI — bring your own key for OpenRouter, Groq, OpenAI, Anthropic, xAI, NVIDIA NIM, or run fully local with Ollama
• Drop-in Fix Prompt — output is a portable Markdown artifact you paste into Claude Code, Cursor, Windsurf, Copilot, or any other agent

---

Install

pip install remy-agent

---

Quick Start

# First run — launches config wizard
remy

# After setup — scan current directory
remy scan

# Deep scan (SAST + LLM logic pass)
remy scan --deep

# Scan a specific path
remy scan /path/to/your/project

# Secrets only (fast, no LLM needed)
remy scan --secrets-only

# Get the Fix Prompt and copy to clipboard
remy prompt --copy

---

Supported Providers

Provider	API Key Required	Notes
OpenRouter	✅ Yes	Routes to 200+ models. Great flexibility.
Groq	✅ Yes	Ultra-low latency. Best for fast bulk scanning.
OpenAI	✅ Yes	GPT-4o and newer models.
Anthropic	✅ Yes	Claude family. Excellent at code reasoning.
xAI	✅ Yes	Grok model family. OpenAI-compatible.
NVIDIA NIM	✅ Yes	Cloud or self-hosted NIM endpoint.
Ollama	❌ No	Fully local. Run ollama serve first.

---

Example Terminal Output

  ██████╗ ███████╗███╗   ███╗██╗   ██╗
  ██╔══██╗██╔════╝████╗ ████║╚██╗ ██╔╝
  ██████╔╝█████╗  ██╔████╔██║ ╚████╔╝
  ██╔══██╗██╔══╝  ██║╚██╔╝██║  ╚██╔╝
  ██║  ██║███████╗██║ ╚═╝ ██║   ██║
  ╚═╝  ╚═╝╚══════╝╚═╝     ╚═╝   ╚═╝

  ┌─────────────────────────────────────────────┐
  │   Remy Scan Report — 14 Finding(s)          │
  │  🔴 CRITICAL  3  ████████░░░░░░░░           │
  │  🟠 HIGH      5  ██████████████░░           │
  │  🟡 MEDIUM    4  ██████████░░░░░░           │
  │  🔵 LOW       2  ████░░░░░░░░░░░░           │
  └─────────────────────────────────────────────┘

---

Example Fix Prompt Output

# Remy Security & Bug Fix Prompt

You are an AI coding agent. Fix the following issues in this codebase.
Apply minimal, targeted changes. Preserve existing behavior and style.

## Findings

### [CRITICAL] Hardcoded production Stripe API key
**File:** `src/services/payments.py`, lines 12–14
**CWE:** CWE-798

**Issue:** A Stripe Live API key was found hardcoded in source code: `sk_l***0xyz`.
Committing secrets to version control is a critical security risk.

**Fix instructions:** Remove from source code immediately. Store in environment
variable: `STRIPE_SECRET_KEY = os.environ['STRIPE_SECRET_KEY']`. Rotate the key
in your Stripe Dashboard.

---

Configuration

Config is stored at ~/.remy/config.toml. API keys are stored in the OS keyring.

Setting	Default	Description
provider	—	LLM provider (openrouter, groq, openai, anthropic, xai, nvidia_nim, ollama)
model	—	Model ID (e.g. gpt-4o, claude-sonnet-4-5)
base_url	provider default	Override base URL (for Ollama or self-hosted NIM)
scan_defaults.deep	false	Enable LLM logic-bug pass by default
scan_defaults.max_file_size_kb	1000	Skip files larger than this
scan_defaults.respect_gitignore	true	Honor .gitignore and .remyignore

Run remy config show to see your current settings (key redacted).

---

Commands

remy                        # First run → wizard; subsequent → scan cwd
remy scan [PATH]            # Full scan
remy scan --deep            # SAST + LLM logic pass
remy scan --secrets-only    # Secrets/credentials only
remy scan --api-surface     # API route audit only
remy scan --bypass-check    # Auth bypass detection only
remy scan --deps            # Dependency CVE scan only
remy scan --format json     # Machine-readable output
remy scan --output FILE     # Write output to file
remy prompt                 # View last scan's Fix Prompt
remy prompt --copy          # Copy Fix Prompt to clipboard
remy config                 # Run config wizard
remy config show            # Print current config
remy config set-provider    # Change provider/model
remy providers list         # List all supported providers
remy version                # Show version

---

Using as a CI Gate

Remy exits with code 2 when Critical or High findings are present, making it easy to use as a CI gate:

# .github/workflows/security.yml
- name: Remy security scan
  run: remy scan . --secrets-only --format json --output scan.json
  # Exits 2 if critical/high findings found

---

Roadmap

• tree-sitter multi-language SAST (JS/TS, Go, Java, Rust, PHP, Ruby)
• VS Code extension for inline findings
• SARIF output format for GitHub Security tab integration
• Custom rule packs via ~/.remy/rules/
• PR diff scanning (scan only changed files)
• HTML report output
• Jira/Linear issue creation from findings

---

Contributing

See CONTRIBUTING.md for setup instructions, how to add SAST rules, and how to add new providers.

---

License

Apache License 2.0 — see LICENSE.

---

Credits

Built by Medusa Security.
Maintained by Abhay Gupta (CTO, Medusa Security).