Skip to content

le2004/agentspec

 
 

Repository files navigation

AgentSpec — Spec-Driven Data Engineering



Claude Code Plugin License: MIT Version

A single AI agent reviewing your data pipeline will miss things.
58 specialized agents with 24 knowledge domains will not.


Install · Quick Start · Commands · Agents · Docs


Why AgentSpec?

Every time you ask an AI to build a data pipeline, it starts from scratch — no memory of partition strategies, no awareness of SCD patterns, no understanding of your data contracts. You get hallucinated SQL, wrong incremental strategies, and pipelines that pass in dev but break in production.

AgentSpec solves this with Spec-Driven Data Engineering: a 5-phase workflow where every phase has access to 24 knowledge base domains, every agent knows its boundaries, and every decision is confidence-scored against real documentation — not guessed.


Install

# Install the plugin (one-time)
claude plugin marketplace add luanmorenommaciel/agentspec
claude plugin install agentspec

Done. Every Claude Code session now has 58 agents, 31 commands, and 24 KB domains. Updates are one command:

claude plugin update agentspec

Override any agent locally — drop a file in .claude/agents/<category>/<agent-name>.md and it takes precedence over the plugin version. See Agent Overrides.

Alternative install methods
# Local testing (no install needed)
git clone https://github.com/luanmorenommaciel/agentspec.git
claude --plugin-dir ./agentspec/plugin

# Legacy copy (pre-plugin, still works)
git clone https://github.com/luanmorenommaciel/agentspec.git
cp -r agentspec/.claude your-project/.claude

Quick Start

Build a data pipeline in 5 phases

/agentspec:brainstorm "Daily orders pipeline from Postgres to Snowflake star schema"
/agentspec:define ORDERS_PIPELINE
/agentspec:design ORDERS_PIPELINE
/agentspec:build ORDERS_PIPELINE
/agentspec:ship ORDERS_PIPELINE

Or jump straight to what you need

/agentspec:schema "Star schema for e-commerce analytics"
/agentspec:pipeline "Daily orders ETL with Airflow"
/agentspec:data-quality models/staging/stg_orders.sql
/agentspec:sql-review models/marts/
/agentspec:data-contract "Contract between orders team and analytics"

Which Command Do I Need?

Data Engineering

I want to... Command Agent
Design a data pipeline / DAG /agentspec:pipeline pipeline-architect
Design a star schema / data model /agentspec:schema schema-designer
Add data quality checks /agentspec:data-quality data-quality-analyst
Optimize slow SQL /agentspec:sql-review sql-optimizer
Choose Iceberg vs Delta Lake /agentspec:lakehouse lakehouse-architect
Build a RAG / embedding pipeline /agentspec:ai-pipeline ai-data-engineer
Create a data contract /agentspec:data-contract data-contracts-engineer
Migrate legacy SSIS / Informatica /agentspec:migrate dbt-specialist + spark-engineer

SDD Workflow

I want to... Command What Happens
Explore an idea /agentspec:brainstorm Compare approaches, discovery questions, YAGNI filter
Capture requirements /agentspec:define Structured requirements with clarity score (min 12/15)
Design architecture /agentspec:design File manifest + pipeline architecture + ADRs
Implement the feature /agentspec:build Auto-delegates to specialist agents per file type
Archive completed work /agentspec:ship Lessons learned + KB updates
Update after changes /agentspec:iterate Cascade-aware updates across all phase documents

Visual & Utilities

I want to... Command
Generate architecture diagrams /agentspec:generate-web-diagram
Create presentation slides /agentspec:generate-slides
Visual implementation plan /agentspec:generate-visual-plan
Review code changes visually /agentspec:diff-review
Review code /agentspec:review
Analyze meeting transcripts /agentspec:meeting
Create a new KB domain /agentspec:create-kb
Share HTML page via Vercel /agentspec:share

How It Works

  BRAINSTORM ──► DEFINE ──► DESIGN ──► BUILD ──► SHIP
  Explore ideas   Scope &    File       Agent      Archive &
  & approaches    contracts  manifest   delegation lessons

                                │
          ┌─────────────────────┼──────────────────────┐
          ▼                     ▼                      ▼
    ┌───────────┐        ┌───────────┐          ┌───────────┐
    │ dbt-spec  │        │ spark-eng │          │ pipeline  │
    │ Models    │        │ Jobs      │          │ DAGs      │
    └─────┬─────┘        └─────┬─────┘          └─────┬─────┘
          └────────────────────┼──────────────────────┘
                               ▼
                         BUILD REPORT
                         Tests + Quality Gates

                          ↻ /iterate
                    Cascade-aware updates

Agent matching: Your DESIGN doc specifies dbt staging models, a PySpark job, and an Airflow DAG — AgentSpec automatically delegates to dbt-specialist, spark-engineer, and pipeline-architect.

Requirements changed? /agentspec:iterate updates any phase document with automatic cascade detection across all downstream docs.


58 Agents Across 8 Categories

Category Count Focus
Architect 8 Schema design, pipeline architecture, medallion layers, GenAI systems
Cloud 10 AWS Lambda, GCP Cloud Run, Supabase, CI/CD, Terraform
Data Engineering 15 dbt, Spark, Airflow, streaming, Lakeflow, SQL optimization
Platform 6 Microsoft Fabric end-to-end (architecture, pipelines, security, AI, logging, CI/CD)
Python 6 Code review, documentation, cleaning, prompt engineering
Workflow 6 Brainstorm, define, design, build, ship, iterate
Dev 4 Codebase exploration, shell scripting, meeting analysis, prompt crafting
Test 3 Test generation, data quality analysis, data contract authoring

Every agent follows the same cognitive framework:

  1. KB-first — check local knowledge base before external sources
  2. Confidence-scored — calculate confidence from evidence, never self-assess
  3. Escalation-aware — transfer to the right specialist when out of domain
  4. Quality-gated — pre-flight checklist before every substantive response

24 Knowledge Base Domains

Category Domains
Core DE dbt · spark · sql-patterns · airflow · streaming
Data Design data-modeling · data-quality · medallion
Infrastructure lakehouse · lakeflow · cloud-platforms · terraform
Cloud aws · gcp · microsoft-fabric · supabase
AI & Modern ai-data-engineering · modern-stack · genai · prompt-engineering
Foundations pydantic · python · testing · shared

Each domain contains an index.md, quick-reference.md, concepts/ (3-6 files), and patterns/ (3-6 files with production code). Agents load domains on-demand, not upfront.


5-Phase Workflow with Quality Gates

Phase Command Output Gate
0. Brainstorm /agentspec:brainstorm BRAINSTORM_{FEATURE}.md 3+ questions, 2+ approaches
1. Define /agentspec:define DEFINE_{FEATURE}.md Clarity Score >= 12/15
2. Design /agentspec:design DESIGN_{FEATURE}.md Complete manifest + schema plan
3. Build /agentspec:build Code + BUILD_REPORT.md All tests pass
4. Ship /agentspec:ship SHIPPED_{DATE}.md Acceptance verified

Project Structure

agentspec/
├── .claude/                 # Source of truth (development)
│   ├── agents/              # 58 agents across 8 categories
│   ├── commands/            # 31 slash commands
│   ├── skills/              # visual-explainer, excalidraw-diagram, agent-router
│   ├── kb/                  # 24 knowledge base domains
│   └── sdd/                 # Templates, contracts, features, archive
│
├── plugin/                  # Distributable Claude Code plugin
│   ├── .claude-plugin/      # Manifest + marketplace config
│   ├── agents/              # Path-rewritten agents
│   ├── skills/              # 5 skills (3 from .claude/ + 2 plugin-only)
│   ├── hooks/               # SessionStart workspace init
│   └── ...                  # commands, kb, sdd, scripts
│
├── plugin-extras/           # Plugin-only content (merged by build)
├── build-plugin.sh          # Packages .claude/ → plugin/
└── docs/                    # Getting started, concepts, tutorials, reference

Documentation

Guide What You'll Learn
Getting Started Install and build your first data pipeline
Core Concepts SDD pillars through a data engineering lens
Tutorials dbt, star schema, data quality, Spark, streaming, RAG
Reference Full catalog: 58 agents, 31 commands, 24 KB domains

Contributing

We welcome contributions. See CONTRIBUTING.md for guidelines.

Agents · KB Domains · Commands · Plugin Development · Documentation


License

MIT — see LICENSE.


About

No description, website, or topics provided.

Resources

License

Contributing

Security policy

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages

  • HTML 65.0%
  • Python 19.4%
  • Shell 14.6%
  • Makefile 1.0%