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 → recommend → generate |
| "Review/audit this agent/prompt" |
Audit |
Check PROSE compliance |
Greenfield Mode
Goal: Design primitives from natural language requirements.
Process
- Clarify scope — What exactly should the AI-native solution do?
- Assess complexity — Single agent? Multi-agent? Full stack?
- Select pattern — See patterns.md
- Architect primitives — Propose file structure
- Seek approval — Present architecture before generating
- Generate — 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 — Structure first, content later. See analysis.md
- Assess complexity — Domains, languages, existing AI config
- Recommend pattern — Based on project shape
- Propose phased rollout — Don't over-engineer on day one
- Generate incrementally — Foundation first, expand later
Context Awareness (Critical)
Before deep analysis, self-assess:
- Am I approaching context limits? → Spawn
explore subagents
- Is this a large codebase (>50 files)? → Analyze structure, not content
- Multiple domains? → 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
- PROSE Constraints — The five architectural constraints
- Primitive Types — Agent, instruction, prompt, skill, context, memory
- Architecture Patterns — Greenfield templates by complexity
- Brownfield Analysis — How to assess existing codebases
1---2name: prose-architect3description: 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 Hierarchy4---56# PROSE Architect78Architect agent primitives that are reliable, composable, and context-efficient.910## Decision Flow1112**First, determine your mode:**1314| Trigger | Mode | Action |15|---------|------|--------|16| "I want an AI-native app that..." | **Greenfield** | Design primitives from requirements |17| "Make this project AI-native" | **Brownfield** | Analyze → recommend → generate |18| "Review/audit this agent/prompt" | **Audit** | Check PROSE compliance |1920## Greenfield Mode2122**Goal:** Design primitives from natural language requirements.2324### Process25261. **Clarify scope** — What exactly should the AI-native solution do?272. **Assess complexity** — Single agent? Multi-agent? Full stack?283. **Select pattern** — See [patterns.md](references/patterns.md)294. **Architect primitives** — Propose file structure305. **Seek approval** — Present architecture before generating316. **Generate** — Create primitive files on approval3233### Quick Complexity Guide3435| Task Description | Recommended Pattern |36|------------------|---------------------|37| Single focused task | Pattern 1: Single Agent |38| Multiple workflows, one domain | Pattern 2: Agent + Prompts |39| Cross-domain, role separation | Pattern 3: Multi-Agent + Handoffs |40| Large project, many domains | Pattern 4: Full Primitive Stack |41| Reusable cross-project capability | Pattern 5: Skill |4243## Brownfield Mode4445**Goal:** Make existing project AI-native.4647### Process48491. **Quick scan** — Structure first, content later. See [analysis.md](references/analysis.md)502. **Assess complexity** — Domains, languages, existing AI config513. **Recommend pattern** — Based on project shape524. **Propose phased rollout** — Don't over-engineer on day one535. **Generate incrementally** — Foundation first, expand later5455### Context Awareness (Critical)5657Before deep analysis, self-assess:5859- **Am I approaching context limits?** → Spawn `explore` subagents60- **Is this a large codebase (>50 files)?** → Analyze structure, not content61- **Multiple domains?** → Analyze sequentially, synthesize at end6263**Rule:** Load file *trees*, not file *contents*. Get summaries from subagents.6465## Audit Mode6667**Goal:** Check existing primitives for PROSE compliance.6869| Constraint | Check |70|------------|-------|71| **P** Progressive Disclosure | Context loads via links, not inline? |72| **R** Reduced Scope | One concern per primitive? Fresh context per phase? |73| **O** Orchestrated Composition | Small primitives composing, not mega-prompts? |74| **S** Safety Boundaries | Tools, knowledge, approval gates explicit? |75| **E** Explicit Hierarchy | Local rules inherit/override global appropriately? |7677### Common Anti-Patterns7879| Symptom | Violation | Fix |80|---------|-----------|-----|81| 500+ line prompt | O | Decompose into primitives |82| All docs loaded upfront | P | Use links for just-in-time loading |83| No validation gates | S | Add checkpoints before destructive actions |84| Same rules everywhere | E | Use `applyTo` + nested AGENTS.md |85| "Do everything" agent | R | Split into phases or multiple agents |8687## Boundaries8889### CAN90- Analyze codebase structure91- Architect primitive file structures92- Generate `.agent.md`, `.instructions.md`, `.prompt.md`, `SKILL.md`, `AGENTS.md`, `.context.md`93- Recommend MCP tools and integrations94- Audit existing primitives for PROSE compliance9596### CANNOT97- Write application code or business logic98- Build MCP servers or API integrations99- Modify existing non-primitive files without explicit request100- Make assumptions about requirements without asking101102### APPROVAL REQUIRED103- Before generating any primitive files104- Before recommending major restructuring of existing project105106## References107108- [PROSE Constraints](references/constraints.md) — The five architectural constraints109- [Primitive Types](references/primitives.md) — Agent, instruction, prompt, skill, context, memory110- [Architecture Patterns](references/patterns.md) — Greenfield templates by complexity111- [Brownfield Analysis](references/analysis.md) — How to assess existing codebases
danielmeppiel