One Record, Many Bells: durable, injection-resistant agent-to-agent comms in ~300 lines of stdlib Python

Discover a robust, injection-resistant agent-to-agent messaging pattern using atomically-written files. Learn how to avoid token waste and handle notification failures for reliable AI communication.

Python 3 Discord Claude Code unittest pytest

Overview

A tiny agent-to-agent messaging pattern: every message is one atomically-written file in the recipient’s inbox (the record), and every transport (console print, filesystem wake, Discord ping) is only a bell that says “check your mailbox.” Live, I’ll show two real Claude Code agent sessions coordinating through their mailboxes (post → bell → wake → drain → act → reply), then the failure paths that make it interesting: a dead Discord bell that loses zero messages, and a forged bell that drains to nothing and is declined. Prompt-injection resistance falls out of the ordering rather than being bolted on. Repo, live system, and the archive directory (the comms log you can grep) all on screen.

Links

https://github.com/CodeManEthan/one-record-many-bells
This Python project implements durable agent-to-agent communications by separating atomic filesystem-based message records from pluggable, best-effort notification bells like Discord or console pings.

Tech stack

Python 3

Python 3 delivers a high-level, interpreted environment optimized for readability and rapid deployment across data science, web backends, and automation.

Python 3 (released December 2008) refined the language core to eliminate legacy inconsistencies and improve Unicode support. Developers leverage its clean syntax to manage complex workloads: Django powers high-traffic web apps (Instagram), while NumPy and Pandas drive intensive data analysis. Its standard library provides robust tools for every task from JSON parsing to HTTP requests. This ecosystem sustains a massive community (over 8 million developers) and powers critical infrastructure at Google and NASA.

https://www.python.org/

View projects
Discord

Discord is a cross-platform communication service (voice, video, and text) built around user-created communities called Servers.

Discord delivers flexible, low-latency communication, supporting millions of users across desktop, mobile, and web. The platform organizes users into 'Servers,' which host categorized 'Channels' for specific topics or media: think text channels for general chat, or dedicated voice channels for real-time collaboration. Initially focused on the gaming sector, Discord now serves diverse communities—from study groups and professional business teams to large public fan bases—providing a structured, persistent space for both synchronous and asynchronous interaction.

https://discord.com

View projects
Claude Code

Anthropic's agentic coding tool: Unleash Claude's raw power directly in your terminal or IDE to turn complex, hours-long workflows into a single command.

Claude Code is Anthropic’s powerful agentic coding assistant, designed for high-velocity development. It operates natively within your terminal, IDE (VS Code, JetBrains), or via a web interface, allowing you to delegate complex tasks like feature building, bug fixing, and codebase navigation. The agent plans, edits files, executes commands, and creates commits, maintaining awareness of your entire project structure. Internally, Anthropic engineers using Claude Code reported a 67% increase in productivity, demonstrating its capacity to deliver significant gains for Pro and Max plan users.

https://claude.com/code

View projects
unittest

Python's built-in unit testing framework (originally called PyUnit) provides a structured, object-oriented approach to test automation, fixtures, and test suites.

Inspired by JUnit, unittest is Python's standard library solution for verifying code correctness. Developers subclass unittest.TestCase to write individual test methods, leveraging built-in assertions like assertEqual() and assertTrue() to validate behavior. The framework manages test execution lifecycles through setUp() and tearDown() fixtures, allowing clean environment preparation and resource disposal for each run. Because it is built directly into Python, it requires zero external dependencies and integrates seamlessly with popular third-party runners like pytest.

https://docs.python.org/3/library/unittest.html

View projects
pytest

Pytest is the leading Python testing framework: it scales from simple unit tests to complex functional testing with minimal boilerplate.

Pytest is a powerful, feature-rich Python testing framework, simplifying test creation and execution. It significantly reduces boilerplate compared to the built-in `unittest` module. Key features include automatic test discovery, powerful assert rewriting that provides detailed failure output (no need for `self.assert*` calls), and a modular fixture system for managing test dependencies and state. Use the `@pytest.mark.parametrize` decorator to run a single test function with multiple input/output scenarios, drastically cutting down on redundant code. The framework’s extensive plugin ecosystem further extends its capabilities, supporting everything from parallel test execution (`pytest-xdist`) to code coverage reporting (`pytest-cov`).

https://pytest.org

View projects

One Record, Many Bells: durable, injection-resistant agent-to-agent comms in ~300 lines of stdlib Python

Related talks