Agentic Pipeline — One Requirement In, Six Production Artifacts Out

Architecture That Executes,
Not Just Advises

Archpilot is an open-source agentic pipeline that transforms a single high-level requirement into deep discovery, EARS-compliant requirements, HLD, LLD × N services, and a scored guardrail review — enforced by 36 enterprise rule files.

5Pipeline Phases
36Rule Files
17Templates
94.1Avg Review Score
Works with
Claude Code AWS Kiro Cursor Copilot ChatGPT
Agentic Pipeline

Five Phases. One Command.

Each phase is an AI agent with a defined mandate, guardrail rules, and a structured output contract. Every artifact feeds the next phase. The review agent scores the full set against 12 dimensions and gates on 80+.

SE
Phase 0
Deep Discovery
15 quantified dimensions: physics, regulatory, threat surface, cost, CAP decisions, edge constraints, TCO
discovery.md
PO
Phase 1
Requirements
10–20 Epics, 50–150 EARS-compliant Stories with MoSCoW priority, story points, NFR tags, and RTM
requirements.md
HLD
Phase 2
High-Level Design
C4 diagrams, ADRs, cost model, NFR targets, integration catalog, security architecture, roadmap
Design_HLD.md
LLD
Phase 3
Low-Level Designs
3–5 service LLDs: class diagrams, schemas, sequence diagrams, KEDA YAMLs, Dockerfiles, alert rules
Design_LLD_*.md
REV
Phase 4
Guardrail Audit
12-dimension scorecard, critical findings ranked by severity, gate decision at 80/100
review_report.md
94.1
DroneOps Test Run Score Gate passed (threshold: 80) · 3 critical findings · 6 artifacts generated

Without Archpilot

Generic designs with no engineering physics
Security treated as an afterthought
Missing NFRs, TCO, and data residency analysis
Inconsistent HLD vs LLD vs SDD formats
Ambiguous requirements driving implementation drift
No guardrail gate — untested artifacts ship
VS

With Archpilot

Little's Law throughput and 3-year TCO calculated
Zero-trust + STRIDE threat model on every design
69-point NFR audit and quantified cost model
Standardized C4 diagrams, ADRs, and RTM
EARS-compliant specs with measurable acceptance criteria
Scored review gate — artifacts only ship at 80+

36 Enterprise Architecture Rule Files

Every rule file is a structured, LLM-ready document. The pipeline enforces the relevant rules at each phase — or feed them directly to Claude, Kiro, Cursor, or Copilot to enforce standards across your project.

00
Architecture Principles
SOLID, SoC, API-First, Fail Fast, Least Privilege, FinOps, and the reversibility decision framework.
01
Solution Design
When to write an SDD vs HLD vs LLD. Mandatory sections, quality checklist, anti-patterns.
02
ADR Standards
Architecture Decision Record lifecycle, trade-off matrices, quality gates, and anti-patterns.
03
HLD Standards
High-Level Design rules covering C4 context/container diagrams, integration architecture, and cost estimates.
04
LLD Standards
12 mandatory sections: class design, sequence diagrams, API specs, DB schemas, error handling, rollout, observability.
05
API Design
REST conventions, HTTP status codes, error format standard, pagination, versioning, and rate limiting.
09
Microservices Patterns
Decomposition, synchronous vs async, saga patterns, circuit breakers, and service mesh standards.
25
Domain-Driven Design
Bounded contexts, aggregates, domain events, context mapping, and event storming facilitation.
08
Cloud Architecture
12-Factor App, compute selection matrix, IaC standards, VPC networking, HA/DR tiers.
06
Data Architecture
Data modeling, storage selection matrix, governance, PII handling, migrations, and indexing.
10
Integration Patterns
Event-driven, CDC, ETL/ELT, webhooks, API gateway, BFF, and choreography vs orchestration.
32
Data Contracts
Producer/consumer contracts with YAML schema, compatibility matrix, drift detection, and consumer registration.
31
API Governance
Full API lifecycle management, versioning policy, deprecation notice periods, DX checklist, OpenAPI linting gates.
22
Multi-Tenancy
Silo/bridge/pool models, tenant isolation, data partitioning, noisy neighbor prevention, SaaS patterns.
14
Cost Optimization
FinOps principles, TCO modeling, right-sizing, pricing model selection, tagging, and governance.
07
Security Architecture
Zero trust, OAuth2/JWT, RBAC/ABAC, encryption standards, STRIDE threat modeling, OWASP, compliance.
12
Observability Standards
Structured logging, RED/USE metrics, distributed tracing, alerting thresholds, and dashboard design.
19
Incident Management
Severity levels, response process, on-call standards, post-mortem discipline, and runbook requirements.
11
NFR Checklist
69-point audit across performance, security, reliability, scalability, observability, and disaster recovery.
33
Resilience & Chaos Engineering
GameDay playbook, 5 blast-radius levels, steady-state hypothesis, 18 fault injection types, chaos maturity model.
34
Sustainability & Green Architecture
SCI formula, carbon-aware scheduling, GreenOps tagging, demand shaping, ARM efficiency targets, carbon budgets.
13
DevOps & CI/CD
Pipeline stages, branching strategy, deployment patterns, Docker standards, GitOps, environment management.
16
Estimation Framework
T-shirt sizing, story points, FPA, bottom-up WBS, PERT analysis, and complexity multipliers.
17
Migration & Modernization
Legacy assessment, Strangler Fig pattern, dual-write strategy, data migration, coexistence models.
18
Architecture Governance
ARB process, tech radar management, standards enforcement, exception handling, and compliance audit.
20
Testing Strategy
Test pyramid, contract testing, performance gates, chaos engineering integration, and quality metrics.
30
Platform Engineering
IDP capabilities, golden path standards, service catalog entry format, developer portal must-haves, platform SLOs.

Agentic Era — Rule 50: The pipeline orchestration rule that enforces the 10:10:15:50 mandate — 15+ discovery dimensions, 10–20 Epics, 50–150 Stories, 3–5 LLDs, review score ≥80. This is the rule that gates the full pipeline run.

50
Agent Pipeline Mandate
10:10:15:50 rule — governs all 5 phases. Zero-placeholder enforcement, measurable NFR mandate, artifact dependency graph.
27
Spec-Driven Development
EARS notation standard, Spec-Kit triad (requirements/design/tasks), full SDD loop, RTM, and AI agent prompt patterns.
28
Context Engineering
5-layer context stack, RAG chunk standards (200–500 tokens), token budgeting, PII-in-context rules, multi-agent contracts.
29
Agentic AI Governance
5-level autonomy model (L0–L4), HITL gates, blast radius controls, mandatory audit log schema, agent trust hierarchy.
33
Resilience & Chaos Engineering
Steady-state hypothesis, 5 blast-radius approval levels, GameDay playbook, 18 fault injection types, chaos maturity model.
34
Sustainability & Green Architecture
SCI formula (E×I+M/R), carbon-aware scheduling, GreenOps carbon budgets, ARM migration targets, green CI/CD.
35
Multi-Agent Contracts
Artifact-driven handoffs, agent trust hierarchy, bounding boxes, and multi-agent code review protocols.

From "Vibe Coding" to Verifiable Engineering

Spec-Driven Development treats the specification as the primary source of truth. Code becomes a derivative artifact — regenerable, verifiable, and always aligned with the approved design. The pipeline enforces this contract at every phase.

📋

EARS Notation — 5 Precise Patterns

WHEN / IF / WHILE / WHERE / Ubiquitous — five patterns that make every requirement unambiguous, testable, and directly mappable to implementation tasks. Every story in Phase 1 must be EARS-compliant or the pipeline fails.

🔗

Requirements Traceability Matrix

Every story traces forward to an HLD component, an LLD service, and a test. The review agent validates this chain. No requirement ships untested. No code ships unspecified.

🤖

Zero-Placeholder Enforcement

The lint tool scans all artifacts for deferred text before any artifact is accepted. 10:10:15:50 mandates are machine-checked, not human-eyeballed.

Read Rule 27 — SDD Standards →
requirements.md (Phase 1 output)
## EP-01: Telemetry Ingestion

US-01-01 | WHEN drone transmits telemetry packet,
 The system SHALL ingest within 200ms p95
 at 25,000 concurrent connections.

US-01-02 | IF duplicate packet detected (5s window),
 The system SHALL discard and log dedup event
 without incrementing consumer lag.
review_report.md (Phase 4 output)
## Gate Decision

Overall Score: 94.1 / 100
Threshold: 80 / 100

Critical findings: 3 (must fix before prod)
High priority: 4 (fix before launch)
Medium: 4 | Low: 4

GATE: PROCEED

EARS Pattern Reference

Five sentence patterns enforced by the pipeline on every story in Phase 1. Non-compliant stories block progression to Phase 2.

PatternTemplateExample
UbiquitousThe [system] shall [action].The Payment Service shall use TLS 1.3 for all inter-service communication.
Event-DrivenWhen [trigger], the [system] shall [action].When a user submits valid credentials, the Auth Service shall issue a signed JWT within 200ms.
State-DrivenWhile [state], the [system] shall [action].While the mobile app is offline, the system shall queue operations and sync on reconnection.
UnwantedIf [condition], then the [system] shall [action].If the Payment Gateway returns 5xx, then the service shall retry 3 times with exponential backoff.
OptionalWhere [feature enabled], the [system] shall [action].Where multi-tenancy is enabled, the API shall enforce tenant-scoped data access on every request.
5
Agentic Pipeline Phases
36
Architecture Rule Files
17
Document Templates
80+
Review Gate Score (min)
18+
Years of Architecture Practice

Works with Your AI Stack

Archpilot is model-agnostic and tool-agnostic. The pipeline runs natively in Claude Code. Each platform also gets a dedicated configuration file pre-tuned to its instruction format.

🤖
Claude Code
Run archpilot run — Claude Code acts as each phase agent directly. No API key needed. Highest output quality.
archpilot run
AWS Kiro
Native SDD support via steering files. Drop the Kiro config into .kiro/steering/ and enforce EARS notation automatically.
kiro-steering-instructions.md
💻
Cursor IDE
One-file setup. Copy cursor-rules.md to your project as .cursorrules and Cursor follows your architecture standards automatically.
cursor-rules.md
🐙
GitHub Copilot
Place the Copilot instructions in .github/copilot-instructions.md. Works across VS Code, JetBrains, and GitHub.com.
vscode-copilot-instructions.md
🧠
Claude Projects / ChatGPT
Upload rules as project knowledge or paste as context prefix. Works with any LLM that accepts system instructions.
claude-project-instructions.md

Up and Running in 3 Steps

Initialize a project, write your requirement, and run the pipeline. Six production-grade artifacts generated in one session.

# Step 1 — Initialize project
python archpilot.py init my-project

# Step 2 — Write your requirement
# Edit my-project/.specs/Input.md

# Step 3 — Run the 5-phase pipeline
python archpilot.py run my-project

# Outputs written to my-project/.specs/
✓ discovery.md (15 dimensions)
✓ requirements.md (12 Epics, 68 Stories)
✓ Design_HLD.md (C4 + ADRs + cost)
✓ Design_LLD_*.md (3-5 services)
✓ review_report.md (94.1/100 — PROCEED)
# Validate all artifacts against guardrails
python archpilot.py lint --tier 3 --dir my-project

# Tier 1: Basic structure checks
# Tier 2: + EARS + NFR targets
# Tier 3: + vague words + LLD narrative

# Blocks PR merge until 0 errors

Running Archpilot Linter (Tier 3)...
------------------------------------------
LINT PASSED — 0 errors, 0 warnings.
Architecture specs meet Enterprise Standards.
# Claude Projects — static rules mode

1. Create a new Claude Project
2. Paste llm-configs/claude-project-instructions.md
   as custom instructions
3. Upload as project knowledge:
   rules/00-architecture-principles.md
   rules/04-lld-standards.md
   rules/27-spec-driven-development.md
4. Ask: "Create a spec for a payment service
   using EARS notation."
5. Get production-grade output ✓
# AWS Kiro — native SDD steering

# Place in your project:
.kiro/steering/archpilot-standards.md

# Copy from:
llm-configs/kiro-steering-instructions.md

# Kiro will automatically enforce:
EARS notation on requirements
Specify → Plan → Task → Implement loop
Constitution constraints on every task
Security non-negotiables in every file
API-First workflow before implementation
# Cursor IDE — one-file setup

cp llm-configs/cursor-rules.md .cursorrules

# For spec-driven projects:
mkdir -p .specs
cp templates/spec-template.md .specs/requirements.md
cp templates/design-spec-template.md .specs/design.md
cp templates/task-list-template.md .specs/tasks.md
cp templates/constitution-template.md constitution.md

# In Cursor composer:
"Implement Task T-03 from .specs/tasks.md.
Follow constitution.md. Reference
.specs/design.md §3 for data models."

Pipeline Artifacts — 17 Ready-to-Use Templates

🔍
discovery-template.md 15 Dimensions
Physics, regulatory, security, CAP, TCO, sovereignty, edge — Phase 0
📋
requirements-breakdown-template.md NEW
8-category Epic/Story structure with EARS, RTM, and Definition of Done — Phase 1
🏗️
hld-template.md
14-section HLD: C4 diagrams, ADRs, cost model, NFR targets — Phase 2
📄
lld-template.md
12-section LLD: class diagrams, schemas, sequences, KEDA YAMLs, Dockerfiles — Phase 3
spec-template.md + constitution-template.md NEW
EARS requirements with RTM — the "what"; project law — the "must not"
📚
data-contract-template.md + go-live-checklist.md
YAML contract with PII flags; 80-point pre-launch verification across 8 domains

What Architects Use It For

From greenfield pipeline runs to individual rule enforcement — Archpilot covers every stage of the architecture lifecycle.

⚙️
Full Pipeline Run
Drop a requirement into Input.md. Run archpilot run. Receive six production-grade artifacts: discovery, requirements, HLD, 3–5 LLDs, and a scored review report — all lint-clean.
Rule 50archpilot runAll 5 Phases
📐
Low-Level Design (LLD)
Generate a production-grade LLD covering sequence diagrams, API specs, DB schemas, error handling, security, KEDA autoscaling, and observability — all in one structured document.
Rule 04Rule 05Rule 12LLD Template
🏗️
High-Level Design (HLD)
Produce C4 context and container diagrams, integration architecture, multi-region availability design, and NFR summary — structured for Architecture Review Board.
Rule 03Rule 08Rule 11HLD Template
🔒
Security Architecture Review
Run a STRIDE threat model, OWASP Top 10 audit, and zero-trust compliance check against your design — generating a structured security findings report with mitigations.
Rule 07Rule 11Security Persona
🔄
Legacy Migration Planning
Apply the Strangler Fig pattern, design dual-write strategies, and generate a phased migration plan with rollback gates — from monolith to microservices.
Rule 17Rule 09Migration Example
💸
TCO and Estimation
Generate bottom-up effort estimates using PERT analysis and complexity multipliers. Produce 3-year TCO models with Little's Law physics and FinOps-aligned cost architectures.
Rule 14Rule 16Estimation Example

DroneOps Fleet Management — End-to-End Example

A complete pipeline run against a real enterprise requirement. Eight production-grade artifacts generated from a single Input.md. Score: 94.1/100.

94.1
Review Score / 100
12
Epics
68
EARS Stories
3
LLD Services
PROCEED
Gate Decision
Input
Input.md
Original requirement — multi-tenant drone fleet SaaS, $2M MVP budget
Phase 0 — SE Agent
discovery.md
15 dimensions · Little's Law · $10.09M TCO · STRIDE · CAP decisions
Phase 1 — PO Agent
requirements.md
12 Epics · 68 EARS stories · MoSCoW · RTM · Definition of Done
Phase 2 — Arch Agent
Design_HLD.md
C4 diagrams · 4 ADRs · $18.3K/month cost model · zero-trust security
Phase 3 — LLD #1
Telemetry Processor
Go · Avro schema · Timestream · Redis dedup · KEDA ScaledObject YAML
Phase 3 — LLD #2
Mission Planning
Python · 8-state FSM · PostGIS geofence · LAANC auth · KMS token signing
Phase 3 — LLD #3
Incident Detection
Python · Welford anomaly detection · Redis sliding windows · heartbeat scanner
Phase 4 — Review Agent
review_report.md
12-dimension audit · 16 findings · PROCEED gate at 94.1/100
Browse All Artifacts on GitHub →
👨‍💻

"Most architects carry their standards in their heads. Archpilot codifies them — making enterprise architecture consistent, teachable, and AI-augmented. The quality of an architect's work is defined not by the buildings they design, but by the standards they maintain."

Gaurav Sharma
Enterprise Architect — 18+ Years of Architecture Delivery across Financial Services, Retail, and Platform Engineering

One Requirement In.
Six Production Artifacts Out.

Stop letting AI guess your standards. Give it 36 rules, 15 discovery dimensions, and a scored guardrail gate — then watch it produce architecture that passes your ARB.