FORK — Evidence Firewall for AI Code Agents

The description states that FORK is a deterministic local verifier designed to run inside an actual Git repository as a Codex plugin. It operates through a staged workflow including:

• Initialization (init)
• Locking the test (lock-test)
• Proposing a repair (propose)
• Human approval (approve)
• Applying the patch (apply)
• Verification (verify)

It uses SHA-256 hashes, Git unified diffs, and strict path parsing to ensure that only approved changes are applied. The system also includes a browser-based viewer that validates evidence artifacts using JSON structures and hash comparisons.

The product is described as being built with TypeScript, React, Node.js, and integrates with OpenAI APIs (specifically GPT-5.6 Sol).

Claim: FORK is a plugin for Codex that enforces deterministic workflows between AI authorship and human approval.

Inferred from the description: The system uses Git-based identity checks, SHA-256 hashes, and structured model outputs to maintain evidence integrity.

The project positions itself as a solution to the credibility gap in AI coding agents, where models can both write tests and fixes but may alter or obscure their own work. The authors state that they wanted something more than an agent that simply says “I fixed it.” Instead, FORK enforces a narrow permission boundary between AI-generated content and human decision-making.

It is described as a code agent firewall, aiming to prevent models from changing the evidence used to judge their own work.

Claim: FORK addresses the problem of AI agents weakening or altering tests they create.

Inferred from the description: The tool introduces a structured, deterministic workflow that separates intelligence (Codex) from authority (human approval).

The description does not explicitly name target customers. However, it implies usage by developers working in Git-based environments, particularly those using AI coding agents like Codex. The tool is described as a plugin for Codex and designed to work within real repositories, suggesting its primary audience includes:

• Developers or teams using AI-assisted development tools
• Organizations concerned with auditability and traceability of code changes

Claim: FORK targets developers who use AI coding agents in Git-based workflows.

Inferred from the description: The tool is built for integration into existing CI/CD pipelines via Codex, suggesting a developer-focused ICP.

There is no mention of pricing or business model in the provided description. No revenue streams, monetization strategies, or customer acquisition plans are described.

Not evidenced

FORK is built using:

• TypeScript
• React 19
• Node.js
• Git integration
• OpenAI API (GPT-5.6 Sol)
• Vite
• SHA-256 hashing
• Deterministic command execution

It uses a staged workflow with independent failure points, and includes:

• A local browser validator
• Structured JSON schema output from GPT-5.6 Sol
• Strict path parsing and Git diff handling
• Verification of patch hashes and test states

Claim: FORK is implemented as a deterministic CLI tool and web viewer with strict validation mechanisms.

Inferred from the description: The system enforces determinism through command contracts, hash identities, and controlled execution paths.

There is no evidence of traction or adoption beyond the hackathon submission. No customers, users, or real-world deployments are mentioned.

Not evidenced

The description does not reference competitors or existing tools in this space. It focuses on the unique value proposition of separating AI intelligence from authority, rather than comparing itself to other solutions.

Not evidenced

• No real-world usage: The tool is only demonstrated in a hackathon setting.
• Limited scope: Designed for use within Codex and Git repositories; unclear if it supports broader CI/CD or IDE integrations.
• Unclear scalability: No indication of how the system would scale beyond a single developer workflow.
• Dependency on Codex: The tool is tightly coupled with Codex, limiting its applicability to other AI coding platforms.

Inferred from the description: The lack of real-world deployment and limited integration points raise concerns about long-term viability.

1. Has FORK been tested or used in any production environments beyond the demo?
2. What are the plans for expanding beyond Codex and Git-based workflows?
3. How does FORK handle edge cases like merge conflicts, complex dependency changes, or multi-file patches?
4. Is there a roadmap for integrating with other AI coding platforms or CI/CD systems?
5. What is the long-term vision for the tool—will it evolve into a commercial product?