dev-pipeline

dev-pipeline

A structured development pipeline for AI coding agents. 8 skills that turn your assistant into a disciplined engineering team.

Inspired by gstack. Built for OpenClaw.

Why

AI coding agents are fast but chaotic. They build features without reviewing them. They ship without testing. They fix symptoms instead of root causes. They say "the code looks good" when it's full of silent failures.

This pipeline adds the discipline that AI agents lack:

THINK → PLAN → BUILD → REVIEW → QA → SHIP

Each stage catches problems that later stages can't:

• Think prevents building the wrong thing
• Plan prevents building it the wrong way
• Review catches bugs that testing misses (silent failures, race conditions, security holes)
• QA catches bugs that code review misses (UI glitches, broken flows, missing error states)
• Ship prevents "it works on my machine" deploys

The 8 Skills

Total: ~1,290 lines across 9 files (8 skills + orchestrator).

Quick Paths

New feature: think → plan → build → review → qa → ship

Bug fix: investigate → build → review → qa → ship

Codebase audit: review → qa → security

Something broke: investigate

Real Results

Built and tested on two iOS apps in one session:

• Mission Control (Next.js dashboard): Found 8 issues, fixed 6, QA'd 4 pages
• DinnerBell (iOS meal planner): Found 12 issues (4 critical), fixed all, shipped Build 20 to TestFlight
• OneLedger (iOS finance): Found 12 issues, fixed 11, shipped Build 77 to TestFlight

Critical bugs caught that manual testing missed:

• App silently served fake data when AI API failed (user had no idea)
• fatalError on database migration = permanent crash loop
• SQL injection via string interpolation in database queries
• Vendor learning system flip-flopping categories across receipt types

Install

Option 1: Clone into OpenClaw skills

cd ~/.openclaw/workspace/skills
git clone https://github.com/DashLabsDev/dev-pipeline.git

Option 2: Copy individual skills

# Just want /review and /investigate?
mkdir -p ~/.openclaw/workspace/skills/dev-review
cp dev-pipeline/skills/review/SKILL.md ~/.openclaw/workspace/skills/dev-review/

mkdir -p ~/.openclaw/workspace/skills/dev-investigate
cp dev-pipeline/skills/investigate/SKILL.md ~/.openclaw/workspace/skills/dev-investigate/

Option 3: Use without OpenClaw

The skill files are just markdown instructions. Feed them to any AI coding assistant (Claude Code, Cursor, Copilot, etc.) as system context.

Key Principles

1. Completeness is cheap with AI. Do the complete thing — the shortcut saves minutes, not hours.
2. Fix first, ask second. Mechanical fixes get applied immediately.
3. Name the file, function, line. Vague findings are worthless.
4. Zero silent failures. Every failure visible to someone.
5. Two-pass review. Critical blocks shipping. Informational informs decisions.
6. No fix without root cause. Understand why before you change code.
7. Cross-model adversarial. Two models finding the same bug = high confidence.
8. Boring by default. Proven tech. Earn your innovation tokens.

Contributing

PRs welcome. Areas that need work:

• Language-specific review rules (Rust, Go, Python, etc.)
• CI/CD integration patterns
• Performance profiling stage
• Accessibility testing checklist
• Better test coverage detection per framework
• Mobile-specific QA checklists (iOS, Android)

License

MIT