agent-native-audit

name	agent-native-audit
description	Run a comprehensive scored audit of agent-native architecture principles
model	claude-sonnet-4.6
platforms	{"copilot":{"model":"gpt-5.3-codex"},"opencode":{"model":"openrouter/moonshotai/kimi-k2.6"}}
argument-hint	[optional: specific principle to audit]
disable-model-invocation	true

Agent-Native Audit

Run a comprehensive review of a codebase against agent-native architecture principles, split the work into scored principle-specific audit tracks, and produce a prioritized summary report.

Announce at start: "I'm using the agent-native-audit skill to run a scored agent-native architecture review."

When to Use

• After adding agent tools, prompts, commands, or shared-workspace behavior
• Before shipping major agent-facing product changes
• When you need a numeric baseline for how agent-native a system really is
• When you want prioritized recommendations instead of a vague architectural opinion

Core Principles to Audit

1. Action Parity - "Whatever the user can do, the agent can do"
2. Tools as Primitives - "Tools provide capability, not behavior"
3. Context Injection - "System prompt includes dynamic context about app state"
4. Shared Workspace - "Agent and user work in the same data space"
5. CRUD Completeness - "Every important entity has full CRUD (Create, Read, Update, Delete)"
6. UI Integration - "Agent actions are reflected in the product experience"
7. Capability Discovery - "Users can discover what the agent can do"
8. Prompt-Native Features - "Features are prompts defining outcomes, not rigid workflow code"

Workflow

Step 1: Load the Local Reference Skill

First, invoke agent-native-architecture to refresh the local vocabulary, principles, checklist, and anti-patterns used in this plugin.

Focus especially on:

• Core principles
• Architecture Review Checklist
• Tool Design
• Files & Workspace
• Context Injection
• UI Integration
• Capability Discovery
• Anti-Patterns

Use the local skill name directly. Do not rely on Every-specific namespaced invocations.

Step 2: Decide Audit Scope

If $ARGUMENTS names a single principle, audit only that principle.

Otherwise, audit all 8 principles.

When auditing a plugin, CLI, or developer-tooling codebase, map "user actions" and "UI" broadly:

• Slash commands
• Prompts
• Agents
• Skills
• CLI entrypoints
• Generated artifacts
• Shared files, docs, and workspace state

Step 3: Launch Parallel Audit Tracks

Launch up to 8 parallel audit tracks using the platform's task/subagent tool.

• On Claude-style platforms, this is typically Task with subagent_type
• On Copilot-style platforms, this is typically task with agent_type

Prefer discovery-oriented agents for evidence gathering and stronger review agents for synthesis when available. In this plugin, agent-native-reviewer, architecture-strategist, explore, and general-purpose are all reasonable choices depending on platform and available agents.

Each audit track should:

1. Enumerate all relevant instances in the codebase
2. Check compliance against the principle
3. Produce a specific numeric score in X/Y format
4. List concrete gaps
5. Recommend the highest-impact fixes first

Audit Track Prompts

Track 1: Action Parity

Audit for ACTION PARITY - "Whatever the user can do, the agent can do."

Tasks:

1. Enumerate all meaningful user actions across product surfaces:
   • UI actions
   • Slash commands
   • CLI actions
   • Form submissions
   • API endpoints
   • File and workspace operations
2. Check which actions have corresponding agent tools, commands, or capabilities
3. Score: "Agent can do X out of Y user actions"

Format:

## Action Parity Audit
### User Actions Found
| Action | Location | Agent Equivalent | Status |
|--------|----------|------------------|--------|
### Score: X/Y (percentage%)
### Missing Agent Capabilities
### Recommendations

Track 2: Tools as Primitives

Audit for TOOLS AS PRIMITIVES - "Tools provide capability, not behavior."

Tasks:

1. Find and read all agent tools, commands, or callable integration surfaces
2. Classify each as:
   • PRIMITIVE (good) - Exposes a capability without embedding decision-making
   • WORKFLOW (risky) - Encodes business logic, orchestration, or branching that should live in prompts
3. Score: "X out of Y tools are proper primitives"

Format:

## Tools as Primitives Audit
### Tool Analysis
| Surface | File | Type | Reasoning |
|---------|------|------|-----------|
### Score: X/Y (percentage%)
### Workflow-Heavy Surfaces
### Recommendations

Track 3: Context Injection

Audit for CONTEXT INJECTION - "System prompt includes dynamic context about app state."

Tasks:

1. Find context injection code, system prompts, and agent setup logic
2. Enumerate what is injected versus what should be injected:
   • Available resources
   • User preferences
   • Recent activity
   • Workspace or session state
   • Available capabilities
   • Relevant history
3. Score: "X out of Y important context categories are injected"

Format:

## Context Injection Audit
### Context Types Analysis
| Context Type | Injected? | Location | Notes |
|--------------|-----------|----------|-------|
### Score: X/Y (percentage%)
### Missing Context
### Recommendations

Track 4: Shared Workspace

Audit for SHARED WORKSPACE - "Agent and user work in the same data space."

Tasks:

1. Identify all important data stores and shared artifacts:
   • Files
   • Databases
   • Documents
   • Session stores
   • Queues
   • Generated outputs
2. Check whether agents and users operate on the same state or separate shadow state
3. Score: "X out of Y important data spaces are truly shared"

Format:

## Shared Workspace Audit
### Data Space Analysis
| Data Space | User Access | Agent Access | Shared? |
|------------|-------------|--------------|---------|
### Score: X/Y (percentage%)
### Isolated State (anti-pattern)
### Recommendations

Track 5: CRUD Completeness

Audit for CRUD COMPLETENESS - "Every important entity has full CRUD."

Tasks:

1. Identify all important entities, resources, or managed objects
2. For each, check whether the agent can:
   • Create
   • Read
   • Update
   • Delete
3. Score per entity and overall

Format:

## CRUD Completeness Audit
### Entity CRUD Analysis
| Entity | Create | Read | Update | Delete | Score |
|--------|--------|------|--------|--------|-------|
### Overall Score: X/Y entities with full CRUD (percentage%)
### Incomplete Entities
### Recommendations

Track 6: UI Integration

Audit for UI INTEGRATION - "Agent actions are reflected in the product experience."

Tasks:

1. Check how agent actions propagate to user-visible surfaces
2. Look for:
   • Immediate UI refresh
   • Shared state updates
   • Eventing or streaming
   • File watching
   • Regeneration steps
   • Clear feedback in CLI or docs
3. Identify silent-action anti-patterns where the agent changes something but the user cannot see it promptly

Format:

## UI Integration Audit
### Agent Action -> User Visibility Analysis
| Agent Action | Surface Update Mechanism | Immediate? | Notes |
|--------------|--------------------------|------------|-------|
### Score: X/Y (percentage%)
### Silent Actions
### Recommendations

Track 7: Capability Discovery

Audit for CAPABILITY DISCOVERY - "Users can discover what the agent can do."

Tasks:

1. Check for these discovery mechanisms:
   • Onboarding or getting-started guidance
   • Help documentation
   • Capability hints in UI or command descriptions
   • Agent self-description
   • Suggested prompts or workflows
   • Empty-state guidance
   • Slash commands or discoverable command lists
2. Score against the mechanisms that are relevant to the product

Format:

## Capability Discovery Audit
### Discovery Mechanism Analysis
| Mechanism | Exists? | Location | Quality |
|-----------|---------|----------|---------|
### Score: X/Y (percentage%)
### Missing Discovery Paths
### Recommendations

Track 8: Prompt-Native Features

Audit for PROMPT-NATIVE FEATURES - "Features are prompts defining outcomes, not code."

Tasks:

1. Read prompts, agents, skills, commands, and orchestration code
2. Classify each important behavior as defined primarily in:
   • PROMPT (good) - Outcome described in natural language
   • CODE (risky) - Business logic hardcoded in conditionals or rigid orchestration
3. Check whether changing behavior usually requires prompt edits or code edits

Format:

## Prompt-Native Features Audit
### Feature Definition Analysis
| Feature | Defined In | Type | Notes |
|---------|------------|------|-------|
### Score: X/Y (percentage%)
### Code-Defined Behaviors
### Recommendations

Step 4: Compile the Summary Report

After all audit tracks finish, compile a single report:

## Agent-Native Architecture Review: [Project Name]

### Overall Score Summary

| Core Principle | Score | Percentage | Status |
|----------------|-------|------------|--------|
| Action Parity | X/Y | Z% | ✅/⚠️/❌ |
| Tools as Primitives | X/Y | Z% | ✅/⚠️/❌ |
| Context Injection | X/Y | Z% | ✅/⚠️/❌ |
| Shared Workspace | X/Y | Z% | ✅/⚠️/❌ |
| CRUD Completeness | X/Y | Z% | ✅/⚠️/❌ |
| UI Integration | X/Y | Z% | ✅/⚠️/❌ |
| Capability Discovery | X/Y | Z% | ✅/⚠️/❌ |
| Prompt-Native Features | X/Y | Z% | ✅/⚠️/❌ |

**Overall Agent-Native Score: X%**

### Status Legend
- ✅ Excellent (80%+)
- ⚠️ Partial (50-79%)
- ❌ Needs Work (<50%)

### Top 10 Recommendations by Impact

| Priority | Action | Principle | Effort |
|----------|--------|-----------|--------|

### What's Working Excellently

[List top 5 strengths]

Success Criteria

• All requested audit tracks complete
• Each principle has a specific numeric score in X/Y format
• Summary table includes all audited principles and status indicators
• Recommendations are prioritized by impact
• Report identifies both strengths and gaps

Optional: Single Principle Audit

If $ARGUMENTS specifies a single principle, run only that audit track and provide detailed findings for that principle.

Valid arguments:

• action parity or 1
• tools or primitives or 2
• context or injection or 3
• shared or workspace or 4
• crud or 5
• ui or integration or 6
• discovery or 7
• prompt or features or 8