agentbible 1.0.0

Language-agnostic correctness checker for scientific numerical code

AgentBible

Language-agnostic correctness checker for scientific numerical code.

SPEC.md is the mathematical source of truth for every implementation in this repository.

Install

Language	Command
Python	pip install agentbible
Rust	cargo add agentbible
Julia	julia -e 'using Pkg; Pkg.add("AgentBible")'
C++	CPMAddPackage("gh:rylanmalarchick/research-code-principles@cpp-v0.1.0")

Python Example

Without validation, a normalization bug can quietly propagate:

import numpy as np


def probabilities(logits: np.ndarray) -> np.ndarray:
    return np.exp(logits)

With AgentBible, the same bug fails at the boundary where it matters:

import numpy as np
from agentbible import validate_finite, validate_normalized_l1, validate_probabilities


@validate_finite
@validate_probabilities
@validate_normalized_l1()
def probabilities(logits: np.ndarray) -> np.ndarray:
    return np.exp(logits)

The invalid output raises immediately instead of silently contaminating later results.

CLI

Validate Python-produced data directly:

bible validate results.npy --check normalized_l1

Inspect a non-Python provenance record through the unified interface:

bible validate --lang rust results.json

Generate a Markdown provenance summary:

bible report results.json

Repository Layout

• agentbible/: Python package, CLI, provenance, context retrieval
• languages/rust/: Rust workspace
• languages/cpp/: header-only C++ library
• languages/julia/: Julia package
• schema/provenance_v1.json: cross-language provenance schema