Open source · MIT licensed

ADLC Skills: the operating model for agentic development

47 skills, 28 commands and 2 review agents for Claude Code and Cowork. They cover what coding-agent frameworks leave out: readiness, agent-ready specs, review capacity, governance, operations and building agents.

$ claude plugin marketplace add shmulikdav/adlc-skills

View on GitHub

Coding agents moved the bottleneck

Agents write code faster than organizations can specify, review and govern it. AI amplifies the system it lands in: strong foundations get faster, weak ones get faster at producing rework.

Specs

Agents fill gaps in vague requirements with plausible guesses.

Review

Agent pull requests wait about five times longer for review than human ones.

Governance

Agents hold real permissions, and skill marketplaces have become a malware vector.

Eight plugins, one per phase

adlc-foundations

Are we ready, how much do we delegate, and is it paying off?

adlc-intent

What exactly should the agent build?

adlc-context

What does the agent need to know about our code and rules?

adlc-build

Which execution framework, and how is the work shaped?

adlc-verify

Is the output right, and can review keep up?

adlc-govern

Is the agent setup safe and compliant?

adlc-operate

How do we release, observe, pay for and learn from agent work?

adlc-agent-engineering

How do we build agents and LLM features as products?

It complements Superpowers, GitHub Spec Kit and Anthropic's official plugins rather than replacing them. See all 47 skills on GitHub

Install only what your role needs

Role	Install	Start with
CTO / VP Engineering	adlc-foundations, adlc-operate	/adlc-assess
Product manager	adlc-intent	/write-agentic-prd
Tech lead	adlc-context, adlc-build, adlc-verify	/fix-review-queue
Platform / DevEx	adlc-operate, adlc-govern	/plan-continuous-ai
Security	adlc-govern	/threat-model
Business leader enabling non-engineers	adlc-govern	/citizen-builder-policy
Team building AI agents	adlc-agent-engineering	/design-agent

Benchmarks show focused skill sets outperform large bundles, so install two or three plugins, not all eight.

Grounded in standards, measured with evals

Every skill names the industry standards it applies and cites at least two primary sources. Every skill ships with eval cases that compare it against a no-plugin baseline using Claude Code's plugin evals. Skills that don't measurably help get fixed or cut.

DORA
Google Engineering Practices
Google SRE
NIST AI RMF
ISO/IEC 42001
OWASP Agentic Top 10
OpenSSF
FinOps for AI
Thoughtworks Technology Radar

First results: adlc-verify

Measured with Claude Code's plugin evals: Claude Haiku answering, Claude Sonnet judging, 3 runs each with and without the plugin. Scores are the share of concrete criteria met.

Skill	Without the kit	With the kit
Definition of done	0.07	0.87
Review capacity	0.05	0.57
Hallucination checks	0.40	0.60

Generic code review showed no gain: the model already reviews code well on its own. Early results on one of eight plugins; full method and caveats on GitHub.

The kit is free. Making it stick is the work.

BrAIght Wave helps engineering organizations move from scattered AI usage to a working agentic development lifecycle: ADLC discovery, team workshops, AI champions programs and governance setup. Our method: Map → Prioritize → Build → Scale.

Book a conversation

FAQ

Is it free?

Yes. MIT licensed, including commercial use.

Does installing it run code on my machine?

No. The plugins contain instructions and agent definitions only: no hooks, servers or binaries.

Does it work outside Claude Code?

The skills follow the open Agent Skills format and work in Cowork, Codex, Cursor and Gemini CLI. Commands, agents and evals are Claude Code features.

How is it different from Superpowers?

Superpowers disciplines the coding session. ADLC Skills covers the operating model around it. They are designed to be used together.

Who maintains it?

Shmulik Davar and BrAIght Wave. Contributions are welcome on GitHub.