SIN-Code Verification Oracle

Automated verification for Python code — security, performance, correctness, style, and test execution. Independent ground-truth for AI-generated code.

Part of the SIN-Code agent-engineering stack. Install all subsystems together via the SIN-Code Bundle.

Features

• Security — hardcoded secrets, unsafe eval, SQL injection (via bandit)
• Performance — nested loops, large allocations, missing memoization (AST heuristics)
• Correctness — property-based smoke tests, compile check (via hypothesis)
• Style — lint, import order, formatting (via ruff)
• Tests — isolated test execution with timeout (via pytest)
• Diagnostics — run compilers, type-checkers, and linters as independent oracles
• Trace diffing — capture behavior fingerprints before and after edits
• Eval harness — run eval suites with a NO-OP baseline for agent benchmarking
• MCP server — expose verification tools to AI agents via the Model Context Protocol

Installation

pip install -e .

Optional MCP server support:

pip install -e ".[mcp]"

See INSTALL.md for detailed setup instructions.

Usage

Library

from sin_code_oracle import VerificationOracle

oracle = VerificationOracle(workspace="/path/to/repo")

# Verify a code snippet
verdict = oracle.verify(code="def foo(): return 42", language="python")
print(verdict.to_json())

# Verify a single file
verdict = oracle.verify_file("src/app.py")

# Verify a directory
verdicts = oracle.verify_directory("src/")

CLI

# Run diagnostics (compilers, type-checkers, linters)
oracle diagnostics

# Full verification
oracle verify --root . --test pytest

# Trace capture and diff
oracle trace_capture "pytest tests/" --out trace.json
oracle trace_diff "pytest tests/" --before trace.json

# Eval suite
oracle eval suite.json --label baseline

# Run MCP server
oracle serve

Testing

pytest tests/ -v

MCP Server

Run the MCP server for agent integration:

# With the CLI
oracle serve

# Or directly
python -m sin_code_oracle.mcp_server

Tools exposed:

• verify_code(code, language="python") — verify code correctness using formal proofs and property-based tests
• generate_properties(code, language="python") — generate property-based tests for given code

Integration

The Oracle is designed to work as part of the SIN-Code ecosystem:

• SIN-Code Bundle — orchestrates all subsystems from a single CLI (sin)
• Intent-Based Diffing (IBD) — trigger re-verification for high-risk changes
• Proof of Correctness (POC) — feed symbolic and property-based proofs into verdicts
• Orchestration — run verification as a gate in CI/agent workflows

License

MIT — see LICENSE.