GitHub – automazeio/ccpm: Project management system for Claude Code using GitHub Issues and Git worktrees for parallel agent execution.

Stop Losing Context: A GitHub-First Collaboration Protocol for AI-Assisted Engineering

This post describes a disciplined system that turns PRDs into epics, epics into GitHub issues, and issues into production code.
This matters because teams still lose context, block on tasks, and ship bugs.
Why should you care?
– It preserves end-to-end traceability.
– It enables parallel, conflict-free work streams.
– It fits the GitHub workflow your team already trusts.

What this system is
This is a collaboration protocol, not another project app.
This protocol uses GitHub Issues as the single source of truth.
That keeps history, labels, and relationships where your team expects them.
Example prompt snippet to start a PRD.
„/pm:new-prd feature-name –title ‚User authentication‘ –summary ‚Vision and success criteria'“.

The 5-phase discipline
Every line of code must trace back to a specification.
No shortcuts.
No assumptions.
Phase 1: Brainstorm and create a PRD with vision, user stories, success criteria, and constraints.
Output: .claude/prds/feature-name.md.
Phase 2: Produce an epic with architecture, technical approach, and dependency mapping.
Output: .claude/epics/feature-name/epic.md.
Phase 3: Break the epic into concrete tasks with acceptance criteria and estimates.
Output: .claude/epics/feature-name/[task].md.
Phase 4: Push epics and tasks to GitHub as issues with labels and relationships.
Phase 5: Specialized agents implement tasks and maintain progress updates and an audit trail.
Tip: Type /pm:help for a concise command summary.

Parallel agents and local context
Most AI-assisted workflows run in isolation today.
That creates silos and lost project state.
This protocol keeps per-epic context in .claude/context/ so agents read and update locally before syncing.
Tasks marked parallel: true enable conflict-free concurrent development.
Your main conversation becomes the conductor, not the orchestra.
For code and examples, see the repository at https://github.com/automazeio/ccpm.

Wrapping up
I recently applied a similar protocol on a cross-functional release.
Context drift stopped being the main blocker.
Parallel work reduced idle time for engineers.
The practical impact for GenAI in product engineering is clear.
You get reproducible traceability, faster cycle time, and clearer audit trails.
That is how generative AI becomes a dependable part of engineering at scale.