automazeio/ccpm

CCPM – The Project Manager Agent

What it is

• An Agent Skill (compatible with the agentskills.io standard) that gives AI coding agents a structured project‑management brain.
• Turns natural language ideas into a full spec‑driven workflow: PRD → Epic → Task → GitHub Issue → Code, with every step recorded in the repository.

Why it matters

• Context never disappears – all specs, decisions, and progress live in plain‑markdown files under .claude/ instead of being lost in chat logs.
• Parallelism – a single issue can be split into several independent streams that run simultaneously on separate agents, dramatically cutting wall‑clock time.
• GitHub‑native – uses GitHub Issues (and optionally the gh-sub-issue extension) as the single source of truth, so teams can collaborate with humans or agents without extra project‑management tools.
• Harness‑agnostic – works with Claude Code, Codex, Factory, Amp, Cursor, OpenCode, etc., simply by pointing the harness at skill/ccpm/.

Key features

Feature	Benefit
Spec‑driven workflow	PRD creation, epic planning, task decomposition are all driven by markdown files, guaranteeing traceability.
Parallel execution system	Marks tasks parallel: true and launches multiple agents that each work in its own git worktree, avoiding conflicts.
GitHub‑first integration	Issues become the project state; comments act as an audit trail. No separate database or Projects board needed.
Deterministic bash scripts	Stand‑up, status, search, next‑task, blocked‑task etc. run instantly as scripts, eliminating extra LLM token cost.
Agent‑skill plug‑and‑play	Install once, then any agents that understand the Agent Skills spec can invoke CCPM with natural language (e.g., “start working on issue 42”).
Full traceability	Every line of code can be traced back to a PRD → Epic → Task → Issue → Commit chain.

Installation

# Clone the repo
git clone https://github.com/automazeio/ccpm.git

# Example for a Factory/Droid harness
ln -s $(pwd)/ccpm/skill/ccpm ~/.factory/skills/ccpm

# Example for Claude Code
ln -s $(pwd)/ccpm/skill/ccpm .claude/skills/ccpm

Prerequisites: git, the GitHub CLI (gh) authenticated, and a GitHub repository where the project will live.

Typical usage flow

1. Brainstorm / PRD – Say “I want to build a notification system” → CCPM asks clarifying questions and writes .claude/prds/notification-system.md.
2. Parse PRD → Epic – “parse the notification‑system PRD” creates .claude/epics/notification-system/epic.md.
3. Decompose – “break down the notification‑system epic” produces task markdown files with metadata (depends_on, parallel).
4. Sync to GitHub – “sync the notification‑system epic to GitHub” creates an epic issue and sub‑issues, renames local task files to the issue numbers, and sets up a dedicated worktree.
5. Execute – “start working on issue 42” launches parallel agents scoped to their own files; each commits with a message like Issue #42: ….
6. Track – “standup” or “what’s blocked” runs the bundled standup.sh/status.sh scripts, printing a real‑time report from the markdown files.

Project structure

skill/ccpm/
├─ SKILL.md            # entry point for the skill
└─ references/
   ├─ plan.md          # PRD → epic logic
   ├─ structure.md     # task decomposition
   ├─ sync.md          # GitHub issue creation
   ├─ execute.md       # parallel launch
   ├─ track.md         # status/standup scripts
   └─ scripts/        # 14 bash helpers (status.sh, standup.sh, …)

.claude/
├─ prds/               # product requirement docs
└─ epics/
   └─ <feature>/       # epic.md + <issue‑id>.md task files

All source‑of‑truth data lives in these plain markdown files.

Proven results (as claimed in the README)

• 89 % less time lost to context switching.
• 5–8 parallel tasks instead of a single serial one.
• 75 % fewer bugs thanks to detailed pre‑coding specs.
• Up to 3× faster feature delivery.
• Internal benchmark shows 100 % success on PRD creation, issue execution, and stand‑up when CCPM is used, versus ~28 % without it.

Who’s behind it

• Developed by Automaze (https://automaze.io), a consultancy focused on AI‑augmented software delivery.
• Lead contact: @aroussi on X (formerly Twitter).

Bottom line CCPM is a lightweight, markdown‑first skill that lets AI coding agents operate like a disciplined project‑manager, preserving context, enabling true parallel work, and keeping everything visible in GitHub Issues. It’s a practical addition for teams that already use LLM‑based coding assistants and want a reproducible, audit‑ready workflow.