Anton Clew — Stop AI From Committing Regret

#devchallenge #githubchallenge #cli #githubcopilot

GitHub Copilot CLI Challenge Submission

This is a submission for the GitHub Copilot CLI Challenge

Anton Clew — Stop AI From Committing Regret

What I Built

Anton Clew is a deterministic, policy-driven CLI that prevents risky or unauthorized file changes before they are committed, adding enforceable safety guardrails to AI-assisted coding workflows in Git repositories.

As AI agents and automation generate changes faster than humans can reliably review them, it becomes easy to accidentally modify sensitive files (keys/secrets), touch critical configuration, or change core code paths without explicit approval. Anton Clew shifts safety left by enforcing policy as code before commits happen.

Developers define simple YAML rules (path and glob-based), and Anton Clew evaluates staged files and classifies each change as:

ALLOW — safe changes that proceed normally
NEEDS_APPROVAL — changes requiring explicit human review
DENY — changes that are blocked entirely

The core design principle is determinism: no probabilistic judgments, no hallucinations—just predictable enforcement that can be trusted in local workflows, pre-commit hooks, and CI pipelines.

How It Works

Anton Clew operates on staged Git changes. Developers define repository policies using a simple YAML file that describes which file paths are allowed, require approval, or are explicitly denied.

When the CLI is run, Anton Clew inspects the staged files and deterministically classifies each change as ALLOW, NEEDS_APPROVAL, or DENY. If any denied or unapproved changes are present, the check fails—preventing risky commits before they land.

This approach enables early, enforceable checks on staged changes, making it practical to integrate Anton Clew into local development workflows, pre-commit hooks, and CI pipelines—especially when AI tools or automation are generating code at high speed.

Demo

GitHub Repo:

https://github.com/earlgreyhot1701D/antonclew

Video Overview (NotebookLM):

https://youtu.be/VDftFjy8KGI

For the demo, I created a small Git repository with a policy file and staged changes designed to trigger all three decision paths. Anton Clew evaluates the staged files and produces clear, deterministic outcomes:

.agentpolicy.yaml and README.md → ALLOW
secrets/private.key → DENY (matches deny rules for sensitive files)
src/example.ts → NEEDS_APPROVAL (matches approval-required paths)

This demonstrates Anton Clew’s intended use case: enforcing guardrails on what is about to be committed, especially when AI tools or automation are generating code at high speed.

My Experience with GitHub Copilot CLI

GitHub Copilot CLI acted as a practical, terminal-native assistant throughout development—particularly during testing, debugging, and build stabilization.

I used Copilot CLI to:

Diagnose failing tests and edge-case mismatches
Fix TypeScript build errors (missing dependencies, implicit any issues)
Resolve Node.js ESM import and runtime issues
Apply focused, minimal patches while keeping changes reviewable
Iterate faster without leaving the terminal or breaking flow

Copilot CLI accelerated problem-solving without replacing judgment. Every change was validated with tests and builds, keeping the project deterministic and reliable.

Credits

This project was built with assistance from GitHub Copilot CLI and ChatGPT, used as development accelerators for debugging, iteration, and problem-solving during the build process.

Final design decisions, validation, testing, and responsibility for the codebase are entirely human.

The name Anton Clew is a playful reference to Son of Anton—a nod to Silicon Valley–style systems thinking, legacy, and the idea of leaving a clear “thread” (a clew) to follow through complex technical problems.

AI assisted. Human approved.

All mistakes are mine.