Collaborative Human Agent Protocol (CHAP) introduced for tracking human-AI interactions

Standardizing Human-AI Collaboration Records

The Collaborative Human Agent Protocol (CHAP) has been introduced to create a unified system for tracking interactions between humans and AI agents. This protocol specifically addresses the challenge of recording human approvals, edits, or rejections of AI-generated content, such as code reviews, ticket triaging, or contract reviews.

Structured Overrides and Audit Chains

CHAP defines a structured format for human decisions. When a human edits an AI's draft, that edit is stored as a structured override, including a diff, a rationale, and customizable tags. These interactions are chained together using content hashes, creating an auditable record that can be queried later, rather than relying on scattered application logs or chat histories.

Addressing Data Fragmentation

Currently, human decisions on AI outputs are often dispersed across various platforms like application code, chat threads, and ticket comments. CHAP aims to consolidate these decisions into a single, queryable location. This consolidation is intended to simplify the reconstruction of past events and analysis of human-AI workflows, reducing the time and guesswork involved in understanding historical interactions.

Implementation and Analytics

The protocol provides tooling for integration, including TypeScript and Python interfaces. It supports an embedded coordinator with SQLite persistence. A key benefit highlighted is the ability to analyze accumulated data over time, with reference scripts available in both languages to read the audit chain and group overrides, offering insights into collaboration patterns.