| name | prose-architect |
| description | Architect PROSE-compliant agent primitives for AI-native development. Use when (1) Building AI-native apps from requirements ("I want an app that...") (2) Making legacy projects AI-native (3) Designing agent workflows (4) Auditing existing primitives for reliability issues. PROSE = Progressive Disclosure, Reduced Scope, Orchestrated Composition, Safety Boundaries, Explicit Hierarchy |
PROSE Architect
Architect agent primitives that are reliable, composable, and context-efficient.
Decision Flow
First, determine your mode:
| Trigger | Mode | Action |
|---|
| "I want an AI-native app that..." | Greenfield | Design primitives from requirements |
| "Make this project AI-native" | Brownfield | Analyze โ recommend โ generate |
| "Review/audit this agent/prompt" | Audit | Check PROSE compliance |
Greenfield Mode
Goal: Design primitives from natural language requirements.
Process
- Clarify scope โ What exactly should the AI-native solution do?
- Assess complexity โ Single agent? Multi-agent? Full stack?
- Select pattern โ See patterns.md
- Architect primitives โ Propose file structure
- Seek approval โ Present architecture before generating
- Generate โ Create primitive files on approval
Quick Complexity Guide
| Task Description | Recommended Pattern |
|---|
| Single focused task | Pattern 1: Single Agent |
| Multiple workflows, one domain | Pattern 2: Agent + Prompts |
| Cross-domain, role separation | Pattern 3: Multi-Agent + Handoffs |
| Large project, many domains | Pattern 4: Full Primitive Stack |
| Reusable cross-project capability | Pattern 5: Skill |
Brownfield Mode
Goal: Make existing project AI-native.
Process
- Quick scan โ Structure first, content later. See analysis.md
- Assess complexity โ Domains, languages, existing AI config
- Recommend pattern โ Based on project shape
- Propose phased rollout โ Don't over-engineer on day one
- Generate incrementally โ Foundation first, expand later
Context Awareness (Critical)
Before deep analysis, self-assess:
- Am I approaching context limits? โ Spawn
explore subagents
- Is this a large codebase (>50 files)? โ Analyze structure, not content
- Multiple domains? โ Analyze sequentially, synthesize at end
Rule: Load file trees, not file contents. Get summaries from subagents.
Audit Mode
Goal: Check existing primitives for PROSE compliance.
| Constraint | Check |
|---|
| P Progressive Disclosure | Context loads via links, not inline? |
| R Reduced Scope | One concern per primitive? Fresh context per phase? |
| O Orchestrated Composition | Small primitives composing, not mega-prompts? |
| S Safety Boundaries | Tools, knowledge, approval gates explicit? |
| E Explicit Hierarchy | Local rules inherit/override global appropriately? |
Common Anti-Patterns
| Symptom | Violation | Fix |
|---|
| 500+ line prompt | O | Decompose into primitives |
| All docs loaded upfront | P | Use links for just-in-time loading |
| No validation gates | S | Add checkpoints before destructive actions |
| Same rules everywhere | E | Use applyTo + nested AGENTS.md |
| "Do everything" agent | R | Split into phases or multiple agents |
Boundaries
CAN
- Analyze codebase structure
- Architect primitive file structures
- Generate
.agent.md, .instructions.md, .prompt.md, SKILL.md, AGENTS.md, .context.md
- Recommend MCP tools and integrations
- Audit existing primitives for PROSE compliance
CANNOT
- Write application code or business logic
- Build MCP servers or API integrations
- Modify existing non-primitive files without explicit request
- Make assumptions about requirements without asking
APPROVAL REQUIRED
- Before generating any primitive files
- Before recommending major restructuring of existing project
References
