You type /addp - and your project is split into parallel tickets. Each agent gets its own branch, writes tests, reviews code. No overwrites, no conflicts, no manual cleanup. The way you always wanted it to work.
An independent community skill for Claude Code - not by Anthropic.
Buy once. Use forever. 14-day money-back guarantee.
You're not alone. Everyone using agents seriously hits these walls.
Two agents, one file. One overwrites the other. You don't notice until nothing builds. An hour of debugging for a mistake that took a second.
Agents commit whatever they want. No checks, no verification. The code looks fine - until it blocks everything in the next session. Then you start over.
One agent at a time. 10 tickets, 10x the wait. You know it should run in parallel - but without isolation, every attempt ends in a merge disaster.
An agent pushed code that doesn't build. Your entire team is blocked. Nobody knows who cleans up. In the end, it's you - as always.
The problem isn't your agents. The problem is that nobody gives them rules. There's a better way.
One command. Everything runs.You type /addp. The skill analyzes your project, creates tickets, launches parallel agents in isolated branches - and merges everything back automatically. Tests, reviews, and safety hooks are built in. You sit back.
Your goal is automatically broken into atomic tasks.
Each task gets its own isolated Git worktree.
Tests must pass before agents get merged.
Write tickets in Markdown format. Define files, dependencies, acceptance criteria.
ADDP creates isolated Git worktrees - each agent gets its own working directory.
Implementation agents work simultaneously in their worktrees. Tests first, then code.
Review agents check each worktree: code quality, tests, acceptance criteria. APPROVED, FIXED or REJECTED.
Topological merge into main. Build + test after each merge. No conflicts thanks to isolation.
All features in every license. You only choose the usage type.
Private, non-commercial use
1 person
one-time
incl. 19% VAT
Update renewal: 32 EUR/year
Commercial / business use
1 seat
one-time
incl. 19% VAT
Update renewal: 72 EUR/year
Commercial / business use
Up to 5 seats
one-time
incl. 19% VAT
Update renewal: 140 EUR/year
Update renewal: 40% of purchase price per year. No subscription, no commitment.
Less than 1 month of Cursor Ultra. Pays for itself in 1 day.
Student or in training? Contact us at - we offer the skill for 49 EUR.
No subscription. Pay once, use forever. 14-day money-back - no risk.
verified_user 14-day money-back guarantee
Everything you need for parallel agent development - ready to use.
Orchestrator logic for Agent Teams + fallback mode
335 lines of formal rules (isolation, TDD, review)
Cross-platform Go binary for safety hooks
Implementation, Review, Research, Plan-Checker (.md with YAML frontmatter)
For use without Agent Teams
11 safety hooks: Guard, NoStash, MergeVerify, BuildCheck, SkipTest, Context Monitor, Injection Guard, Secret Scan + lifecycle hooks
macOS: signed .pkg installer. Windows/Linux: install.bat or install.sh
Standard format for structured tickets
Project-specific configuration
addp-hook worktree status - JSON/table report for all worktrees (commits, build, tests, conflicts)
More security, better planning, smart cost optimization.
Automatically detects when agents approach context limits and warns in time.
Prompt injection guard and secret scanning protect your code from accidental leaks.
Opus for planning, Sonnet for code, Haiku for research - automatic cost optimization.
8-dimensional check BEFORE implementation - no wasted context.
Review checks not just code quality, but whether the ticket goals were actually achieved.
Live status line with context bar, model and current ticket.
When an agent fails: automatic retry, decompose or escalation.
Typo fixes and config changes without sprint overhead.
Worktree isolation, TDD gate, mandatory review
Cross-platform Go binary, no Python/Bash dependency
Real Agent Teams (TeamCreate, SendMessage, autonomous teammates)
Web research, parallel research agent, TeamDelete fix
Commit enforcement hooks, merge guard fix, broadcast fix
60+ write vector detection, cross-platform guard (Win/Mac/Linux)
BuildCheck, SkipTest detector, worktree status, offline PDF docs
Context Monitor, security suite, model profiles, plan verification, quick mode
Test for 14 days. If the setup doesn't work or the skill doesn't fit your workflow - full refund, no questions asked.
ADDP supports Go, TypeScript, Python, Rust and C++ out-of-the-box. Configuration is language-agnostic - you can define build, test and lint commands in addp.toml.
Under 5 minutes. On macOS: double-click the signed .pkg installer. On Windows/Linux: run install.bat or install.sh. Then you can run /addp in Claude Code right away.
No. ADDP has two modes: Agent Teams (when CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 is set) and a subagent fallback mode that works without Agent Teams.
You keep everything you downloaded - permanently. After 12 months, only updates end. You can buy an update renewal at any time (40% of the original price per year).
Enter your license key on the downloads page to receive a new download link. The link is valid for 72 hours. You can find your license key in your confirmation email.
With the Private license you can use the skill on one machine. For commercial use there's the Commercial license (1 seat) and the Team license (up to 5 developers). Larger teams can contact us for a custom offer.
Yes, you'll automatically receive an invoice by email. For B2B customers with a VAT ID, we issue invoices with reverse charge notation for EU cross-border transactions.
14 days from purchase. If the skill doesn't work or doesn't fit your workflow, we refund the full purchase price. No questions, no conditions.
Yes, the skill files are Markdown and therefore fully readable and customizable. The hook binary (addp-hook) is compiled Go. You can adapt the skill to your project.
No. ADDP is an independent community skill developed by kobosil.me Development. We use Claude Code's open skill architecture but are not affiliated with Anthropic.
We recommend Claude Max ($100/month) since ADDP runs multiple agents in parallel, which uses up the standard quota quickly. Max gives you 5x more capacity and Opus 4.6 - ideal for parallel sprints.
Every day. ADDP isn't a theory product - it was built from real daily work. Every feature exists because we needed it ourselves. The entire development of ADDP runs on ADDP.
kobosil.me Development, Berlin. Over 10 years of software architecture experience - Go, TypeScript, cloud infrastructure. Not a marketing startup, but developers who use this tool every single day.
Yes. Other vendors sell you 1,700+ generic prompt collections and call it an "operating system." ADDP is a single, well-engineered skill - with real agent workflows, safety hooks, and mandatory reviews. One system that works beats a thousand templates that don't.
You don't need 1,792 skills. You need one that works.
ADDP is the original - not an inflated skill collection, but a working system built from real daily work. Worktree isolation, TDD gate, mandatory reviews, safety hooks - all built because we need it ourselves every day.
Built by kobosil.me Development - 10+ years of software architecture.
ADDP - Agent-Driven Development Process. In production since 2025. The original.