Architect

Decompose large system specifications into epic beads using DDD bounded contexts.

1|Updated Apr 25, 2026
One-click install
npx skills add https://github.com/htxryan/claude-code-config-profiles --skill architect-htxryan
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: Architect
Source: https://github.com/htxryan/claude-code-config-profiles/tree/main/.agents/skills/compound/architect
Command: npx skills add https://github.com/htxryan/claude-code-config-profiles --skill architect-htxryan

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Breaking a large system specification into implementation-ready work items is error-prone: boundaries get drawn wrong, interface contracts go undocumented, and cross-epic integration failures surface too late. This Skill turns a system spec into naturally-scoped epic beads with explicit contracts, dependency wiring, and human approval gates. ## Core Features & Use Cases - Six-phase decomposition pipeline: Socratic domain discovery with a research sufficiency gate, system-level EARS specification, six-parallel-subagent decomposition (bounded contexts, dependencies, sizing, interfaces, STPA hazards, gap analysis), and bead materialization with dependency wiring. - Advisory fleet: Solicits independent architectural perspectives from external model CLIs (Claude, Gemini, Codex) across security, scalability, organizational, and simplicity lenses before spec approval. - Integration Verification epic: Automatically creates a final IV epic scoped LIGHT/MEDIUM/FULL based on contract classification, with a contracts-under-test table. - Optional loop launch: Opt-in phases hand off materialized epics to the infinity loop and polish loop for autonomous implementation. - Use Case: Given a 50-page system spec for a new platform, produce 6-10 epic beads each with scope boundaries, EARS subsets, explicit and implicit interface contracts, and wired dependencies, ready for autonomous cook-it processing. ## Quick Start Ask the AI to decompose your system specification into epic beads by providing a beads epic ID or a markdown spec file path.

Frequently Asked Questions about Architect

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I decompose a large system spec into epics?▼

Provide a beads epic ID or a markdown file path as input. The skill runs six phases: Socratic domain discovery, system-level EARS spec writing, six-subagent decomposition using DDD bounded contexts, bead materialization with dependency wiring, and optional loop launch.

What is a bounded context in domain-driven design decomposition?▼

A bounded context is a natural domain boundary with a coherent ubiquitous language. The skill uses a bounded context mapper subagent to propose candidate epics, then validates them against structural, semantic, organizational, and economic criteria before approval.

Does the Architect skill require human approval during decomposition?▼

Yes, three human gates are mandatory: Gate 1 confirms domain understanding, Gate 2 approves the system spec with the advisory fleet brief, and Gate 3 approves the epic structure and interface contracts. Gates 4 and 5 activate only with opt-in launch phases.

Can I use external AI models to review my architecture spec?▼

Yes, the advisory fleet detects available CLIs (Claude, Gemini, Codex) and spawns them in parallel across four lenses: security, scalability, organizational delivery, and simplicity. Feedback is synthesized into a brief persisted to docs/specs/ and shown at Gate 2.

Why do cross-epic integration failures appear late in development?▼

Without explicit integration testing, interface mismatches between epics only surface at the end. The skill creates an Integration Verification epic that depends on all domain epics and tests cross-epic contracts scoped LIGHT, MEDIUM, or FULL by contract type.

When should I not use the infinity loop launch phase?▼

Skip Phase 5 if you only need the decomposition artifacts. The launch phase is opt-in and must go through the launch-loop command, never direct ca loop invocation. Always dry-run first and launch in a screen session.