The Agent Skill That Replaces Vibe Coding with GitHub Issues

What it does CCPM gives compatible AI harnesses—Claude Code, Cursor, Codex, and others—a structured project-management brain. It enforces a five-phase workflow from brainstorming and PRD writing through task decomposition, GitHub issue sync, and parallel execution, storing all state in local markdown files inside your repo. The goal is to replace context loss and ad-hoc “vibe coding” with persistent, traceable specifications.

The interesting bit Instead of treating a GitHub issue as a single serial task, CCPM analyzes it for independent work streams—say, database migrations versus API endpoints—and spins up multiple agents in the same Git worktree simultaneously. Routine reporting like standup and status is handled by bash scripts rather than the LLM, so answers are instant and token-free.

Key highlights

• Uses GitHub Issues as the single source of truth; no external project-management database.
• Parallel execution via Git worktrees lets several agents tackle one epic concurrently.
• All project state lives in .claude/ markdown files, making context fully portable across sessions.
• Deterministic ops (status, standup, search) are handled by 14 bash scripts, not the LLM.
• Works with any Agent Skills–compatible harness; no lock-in to a single editor or model.

Caveats

• Requires a GitHub repository and authenticated gh CLI; the entire workflow is built around GitHub Issues.
• The rigid five-phase discipline may feel heavy for quick experiments or prototypes.

Verdict Worth a look if you are coordinating multiple AI agents on real features and need to keep them from colliding. Skip it if you just want a lightweight chat-to-code scratchpad.