Skip to content
UXClaim
Design Ops

Autonomous Development Pipeline

Claude Code skill turning specs into shipped code through adaptive phases with feedforward guides, feedback sensors, and requirement traceability.

View source More in Design Ops

What It Does

ADP is a Claude Code skill that transforms feature specifications into committed, validated code through four adaptive phases: Specify → Design → Tasks → Execute. It intelligently routes work based on complexity—quick mode for small changes, full pipeline for complex features—ensuring spec-to-commit traceability at every step.

How It Works

ADP operates on two layers: a Skill layer defining the methodology and a Runtime layer providing TypeScript helpers. The pipeline starts with adp init to detect your stack and generate feedforward guides from your codebase. Then adp run <feature> executes the full flow, with feedback sensors (typecheck, lint, tests) enforced at phase boundaries. State persists between sessions—pause with adp pause, resume with adp resume.

Key Features

Feedforward Guides: Automatically captures architecture, conventions, testing patterns, and risks before design begins. Sensor Gates: Runs validation at each phase boundary; adaptive retry for resilience. Traceability: Requirements, tasks, and commits stay linked by stable IDs. Agent Operations: Live TUI dashboard, evaluator agents, stuck detection, gated push/PR flow. Auto-Mode: Maximum autonomy variant with adaptive retry and one-click approval gates.

Who Benefits

Design leaders automating specification-to-code workflows, product managers coordinating complex features, and teams needing audit trails of design decisions linked to shipped code.

Frequently asked questions

How do I install ADP?
One-line install via curl (macOS/Linux) or PowerShell (Windows). Run `curl -fsSL https://raw.githubusercontent.com/0xPuncker/adp/main/bin/install.sh | bash` to install globally. Use `ADP_SKILL_ONLY=1` to skip CLI setup. Requires git, curl, and Node.js ≥22 for CLI mode.
What does ADP actually do?
ADP automates spec-to-code workflows. Feed it a feature request; it clarifies requirements (Specify), generates design and tasks, executes builds, runs sensors (lint/test/typecheck), and commits with traceability. Phases adapt based on complexity—small changes skip to execution; complex features run full pipeline.
How does ADP track requirements to commits?
ADP maintains an ID chain from requirements through specs, tasks, and commits. Each requirement gets a stable ID (REQ-01.2); tasks inherit that ID; commits reference task IDs. This ensures every shipped line of code traces back to the original specification.
Can I pause and resume work?
Yes. Run `adp pause` to snapshot your progress to HANDOFF.md and stop. Later, run `adp resume` to read the handoff state and continue exactly where you left off. State persists across sessions automatically.
What is auto-mode and when should I use it?
Auto-mode (`adp auto-mode <feature>`) runs maximum autonomy: skips clarification steps, retries sensor failures up to 3 times, and gates push/PR with one-click approval. Use it for well-scoped, straightforward features. It halts on persistent sensor failures or denied gates.
What are feedforward guides?
Guides are `.adp/guides/` files auto-generated from your codebase—capturing stack, architecture, conventions, testing, integrations, and risks. Claude reads these before design starts, eliminating context-building overhead and ensuring generated code matches your patterns.
Does ADP work without Node.js?
Yes. Use `ADP_SKILL_ONLY=1` during install to get skill files only (no CLI). Or build a standalone binary with `npm run build:standalone` for air-gapped environments. Both modes work with Claude Code; CLI adds convenience.
What sensors does ADP run?
By default: lint, typecheck, unit tests. You define custom sensors in `harness.yaml`. Sensors run at phase boundaries and before every commit. If a sensor fails, ADP retries or halts; in auto-mode, it attempts adaptive fixes up to 3 times.

Glossary

Feedforward Guide
Auto-generated documentation from your codebase capturing architecture, naming conventions, test patterns, and integration points. Claude reads these before designing to ensure generated code fits your project's style and structure.
Sensor
Validation tool (typecheck, lint, test, custom harness command) that runs at phase boundaries and before commits. Sensors enforce quality gates; failures trigger adaptive retry or halt execution.
ID Chain
Stable identifier linking specifications (REQ-01.2) through design, tasks, commits, and validation artifacts. Enables end-to-end traceability: which spec drove which task, which task drove which commit.
Harness
YAML config defining your project's tech stack, testing frameworks, linting rules, custom validation commands, and agent autonomy thresholds. Generated by `adp init`, customized per project.
Auto-Mode
Unattended execution variant that maximizes agent autonomy—skips clarification, applies adaptive retry on sensor/evaluator failures, and gates final push/PR for one-click approval before shipping.

More in Design Ops

All →
Design Ops

Ibras AI Agent

Claude Code marketplace for building and evaluating AI agents. Install plugins to generate complete Mastra evaluation systems with scorers, datasets, experim...

volfadar
Design Ops

Claude Skills

Production-grade Claude AI skills for stock analysis, prompt engineering, meeting documentation, and UX design with visual-first outputs.

bhrpraju
Design Ops

Propose

Run three parallel design agents with different philosophies to generate ranked approach proposals for any design decision.

brianharms
Design Ops

SOTA Present

Generate polished HTML slides, editable PowerPoint, and Feishu whiteboards from one content description with coordinated design and anti-slop taste rules.