Agent-Native Dev Protocols

Research project investigating a single question:

Are we still just hacking our way through AI-agent software development with markdown files — or is anyone building an actual agent-native protocol and standard covering the lifecycle from spec definition through task tracking?

That is: a protocol designed from the ground up for AI coding agents as the primary consumers and producers of specs, task graphs, and progress state — where human readability is a secondary concern — rather than human project-management tooling (Jira, Linear, sprints) retrofitted with agent integrations.

Repository structure

Path	Contents
research-question/	The research question — refined framing plus the original verbatim capture
research/	Gathered findings and source notes from the deep-research pass
report/	Typst source (report.typ) and the compiled PDF report

The report

The final write-up is compiled with Typst and published as a PDF attached to the latest GitHub release — see the Releases page of this repository.

Method

1. Capture and refine the research question
2. Multi-angle web research with adversarial claim verification (Claude Code deep-research workflow)
3. Synthesis into research/ findings
4. Report authored in Typst, compiled to PDF
5. Published as a GitHub release

Researched and written with Claude Code (Claude Fable 5), July 2026.