@vibe-agent-toolkit/vat-development-agents 0.1.8 → 0.1.11

package/README.md

@@ -46,10 +46,40 @@ Identifies opportunities to improve agent resources following Anthropic's "small
 
 ## Installation
 
+### As NPM Package
+
 ```bash
 npm install @vibe-agent-toolkit/vat-development-agents
 ```
 
+### Installing the Skill
+
+This package includes a distributable Claude Code skill called `vibe-agent-toolkit` that provides guidance on using VAT for agent creation and development.
+
+**Install from npm:**
+```bash
+vat skills install npm:@vibe-agent-toolkit/vat-development-agents
+```
+
+**Install from local directory (development):**
+```bash
+vat skills install ./packages/vat-development-agents
+```
+
+**Verify installation:**
+```bash
+vat skills list --installed
+```
+
+The `vibe-agent-toolkit` skill will be installed to `~/.claude/plugins/vibe-agent-toolkit/` and will appear in Claude Code after restarting or running `/reload-skills`.
+
+**What the skill includes:**
+- VAT overview and use cases
+- Agent creation guide (using agent-generator)
+- CLI workflow documentation
+- Links to agent authoring and orchestration guides
+- Best practices for portable agents
+
 ## Usage
 
 ### As NPM Package

package/dist/agents/pattern-analyzer.d.ts

@@ -0,0 +1,34 @@
+/**
+ * Pattern Analyzer Agent
+ * LLM Analyzer that identifies appropriate agent archetypes
+ */
+export interface PatternAnalyzerConfig {
+    name: string;
+    version: string;
+}
+export interface DesignRequest {
+    agentPurpose: string;
+    successCriteria: string[];
+    constraints?: string[];
+}
+export interface PatternAnalysis {
+    recommendedArchetype: 'pure-function-tool' | 'llm-analyzer' | 'conversational-assistant' | 'agentic-workflow';
+    confidence: 'high' | 'medium' | 'low';
+    reasoning: string;
+    recommendedLLM?: string;
+    alternativePatterns?: string[];
+}
+export interface PatternAnalyzerAgent {
+    metadata: {
+        name: string;
+        version: string;
+        archetype: 'llm-analyzer';
+    };
+    systemPrompt: string;
+    analyze: (request: DesignRequest) => Promise<PatternAnalysis>;
+}
+/**
+ * Create a pattern analyzer agent with resource-based prompts
+ */
+export declare function createPatternAnalyzer(config: PatternAnalyzerConfig): PatternAnalyzerAgent;
+//# sourceMappingURL=pattern-analyzer.d.ts.map

package/dist/agents/pattern-analyzer.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"pattern-analyzer.d.ts","sourceRoot":"","sources":["../../src/agents/pattern-analyzer.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAMH,MAAM,WAAW,qBAAqB;IACpC,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,WAAW,aAAa;IAC5B,YAAY,EAAE,MAAM,CAAC;IACrB,eAAe,EAAE,MAAM,EAAE,CAAC;IAC1B,WAAW,CAAC,EAAE,MAAM,EAAE,CAAC;CACxB;AAED,MAAM,WAAW,eAAe;IAC9B,oBAAoB,EAAE,oBAAoB,GAAG,cAAc,GAAG,0BAA0B,GAAG,kBAAkB,CAAC;IAC9G,UAAU,EAAE,MAAM,GAAG,QAAQ,GAAG,KAAK,CAAC;IACtC,SAAS,EAAE,MAAM,CAAC;IAClB,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,mBAAmB,CAAC,EAAE,MAAM,EAAE,CAAC;CAChC;AAED,MAAM,WAAW,oBAAoB;IACnC,QAAQ,EAAE;QACR,IAAI,EAAE,MAAM,CAAC;QACb,OAAO,EAAE,MAAM,CAAC;QAChB,SAAS,EAAE,cAAc,CAAC;KAC3B,CAAC;IACF,YAAY,EAAE,MAAM,CAAC;IACrB,OAAO,EAAE,CAAC,OAAO,EAAE,aAAa,KAAK,OAAO,CAAC,eAAe,CAAC,CAAC;CAC/D;AAED;;GAEG;AACH,wBAAgB,qBAAqB,CAAC,MAAM,EAAE,qBAAqB,GAAG,oBAAoB,CAiDzF"}

package/dist/agents/pattern-analyzer.js

@@ -0,0 +1,71 @@
+/**
+ * Pattern Analyzer Agent
+ * LLM Analyzer that identifies appropriate agent archetypes
+ */
+import * as CorePrinciples from '../../generated/resources/prompts/core-principles.js';
+import * as LLMSelection from '../../generated/resources/prompts/llm-selection.js';
+import * as PatternRecognition from '../../generated/resources/prompts/pattern-recognition.js';
+/**
+ * Create a pattern analyzer agent with resource-based prompts
+ */
+export function createPatternAnalyzer(config) {
+    // Build system prompt from fragments
+    const systemPrompt = `
+You are an expert agent architect that analyzes requirements and recommends appropriate agent patterns.
+
+${CorePrinciples.fragments.purposeDriven.text}
+
+${CorePrinciples.fragments.simpleFirst.text}
+
+${PatternRecognition.text}
+
+${LLMSelection.text}
+
+Your job: Analyze the design request and recommend the best agent archetype and LLM.
+
+Output JSON matching this schema:
+{
+  "recommendedArchetype": "pure-function-tool | llm-analyzer | conversational-assistant | agentic-workflow",
+  "confidence": "high | medium | low",
+  "reasoning": "string - explain why this pattern fits",
+  "recommendedLLM": "string - specific model recommendation",
+  "alternativePatterns": ["string array - other viable options"]
+}
+  `.trim();
+    return {
+        metadata: {
+            name: config.name,
+            version: config.version,
+            archetype: 'llm-analyzer',
+        },
+        systemPrompt,
+        /**
+         * Analyze a design request and recommend pattern
+         */
+        async analyze(_request) {
+            // This would integrate with an actual LLM in a real implementation
+            // For now, demonstrate the structure
+            return {
+                recommendedArchetype: 'conversational-assistant',
+                confidence: 'high',
+                reasoning: 'Multi-turn dialogue with context accumulation',
+                recommendedLLM: 'claude-3-5-sonnet-20241022',
+                alternativePatterns: ['llm-analyzer'],
+            };
+        },
+    };
+}
+// Example usage demonstrating resource composition
+const analyzer = createPatternAnalyzer({
+    name: 'pattern-analyzer',
+    version: '1.0.0',
+});
+console.log('Pattern Analyzer Agent:');
+console.log(`- System prompt length: ${analyzer.systemPrompt.length} chars`);
+console.log(`- Resources used: ${Object.keys(CorePrinciples.fragments).length + Object.keys(PatternRecognition.fragments).length + Object.keys(LLMSelection.fragments).length} fragments`);
+console.log(`- Archetype: ${analyzer.metadata.archetype}`);
+console.log('\nPrompt composition:');
+console.log(`  • Core principles: ${Object.keys(CorePrinciples.fragments).length} fragments`);
+console.log(`  • Pattern recognition: ${Object.keys(PatternRecognition.fragments).length} fragments`);
+console.log(`  • LLM selection: ${Object.keys(LLMSelection.fragments).length} fragments`);
+//# sourceMappingURL=pattern-analyzer.js.map

package/dist/agents/pattern-analyzer.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"pattern-analyzer.js","sourceRoot":"","sources":["../../src/agents/pattern-analyzer.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,cAAc,MAAM,sDAAsD,CAAC;AACvF,OAAO,KAAK,YAAY,MAAM,oDAAoD,CAAC;AACnF,OAAO,KAAK,kBAAkB,MAAM,0DAA0D,CAAC;AA+B/F;;GAEG;AACH,MAAM,UAAU,qBAAqB,CAAC,MAA6B;IACjE,qCAAqC;IACrC,MAAM,YAAY,GAAG;;;EAGrB,cAAc,CAAC,SAAS,CAAC,aAAa,CAAC,IAAI;;EAE3C,cAAc,CAAC,SAAS,CAAC,WAAW,CAAC,IAAI;;EAEzC,kBAAkB,CAAC,IAAI;;EAEvB,YAAY,CAAC,IAAI;;;;;;;;;;;;GAYhB,CAAC,IAAI,EAAE,CAAC;IAET,OAAO;QACL,QAAQ,EAAE;YACR,IAAI,EAAE,MAAM,CAAC,IAAI;YACjB,OAAO,EAAE,MAAM,CAAC,OAAO;YACvB,SAAS,EAAE,cAAuB;SACnC;QACD,YAAY;QAEZ;;WAEG;QACH,KAAK,CAAC,OAAO,CAAC,QAAuB;YACnC,mEAAmE;YACnE,qCAAqC;YAErC,OAAO;gBACL,oBAAoB,EAAE,0BAA0B;gBAChD,UAAU,EAAE,MAAM;gBAClB,SAAS,EAAE,+CAA+C;gBAC1D,cAAc,EAAE,4BAA4B;gBAC5C,mBAAmB,EAAE,CAAC,cAAc,CAAC;aACtC,CAAC;QACJ,CAAC;KACF,CAAC;AACJ,CAAC;AAED,mDAAmD;AACnD,MAAM,QAAQ,GAAG,qBAAqB,CAAC;IACrC,IAAI,EAAE,kBAAkB;IACxB,OAAO,EAAE,OAAO;CACjB,CAAC,CAAC;AAEH,OAAO,CAAC,GAAG,CAAC,yBAAyB,CAAC,CAAC;AACvC,OAAO,CAAC,GAAG,CAAC,2BAA2B,QAAQ,CAAC,YAAY,CAAC,MAAM,QAAQ,CAAC,CAAC;AAC7E,OAAO,CAAC,GAAG,CAAC,qBAAqB,MAAM,CAAC,IAAI,CAAC,cAAc,CAAC,SAAS,CAAC,CAAC,MAAM,GAAG,MAAM,CAAC,IAAI,CAAC,kBAAkB,CAAC,SAAS,CAAC,CAAC,MAAM,GAAG,MAAM,CAAC,IAAI,CAAC,YAAY,CAAC,SAAS,CAAC,CAAC,MAAM,YAAY,CAAC,CAAC;AAC3L,OAAO,CAAC,GAAG,CAAC,gBAAgB,QAAQ,CAAC,QAAQ,CAAC,SAAS,EAAE,CAAC,CAAC;AAC3D,OAAO,CAAC,GAAG,CAAC,uBAAuB,CAAC,CAAC;AACrC,OAAO,CAAC,GAAG,CAAC,wBAAwB,MAAM,CAAC,IAAI,CAAC,cAAc,CAAC,SAAS,CAAC,CAAC,MAAM,YAAY,CAAC,CAAC;AAC9F,OAAO,CAAC,GAAG,CAAC,4BAA4B,MAAM,CAAC,IAAI,CAAC,kBAAkB,CAAC,SAAS,CAAC,CAAC,MAAM,YAAY,CAAC,CAAC;AACtG,OAAO,CAAC,GAAG,CAAC,sBAAsB,MAAM,CAAC,IAAI,CAAC,YAAY,CAAC,SAAS,CAAC,CAAC,MAAM,YAAY,CAAC,CAAC"}

package/dist/agents/requirement-gatherer.d.ts

@@ -0,0 +1,30 @@
+/**
+ * Requirement Gatherer Agent
+ * Conversational assistant that extracts agent design requirements
+ */
+export interface RequirementGathererConfig {
+    name: string;
+    version: string;
+}
+export interface DesignRequest {
+    agentPurpose: string;
+    successCriteria: string[];
+    constraints?: string[];
+}
+export interface RequirementGathererAgent {
+    metadata: {
+        name: string;
+        version: string;
+        archetype: 'conversational-assistant';
+    };
+    systemPrompt: string;
+    processMessage: (message: string) => Promise<{
+        response: string;
+        extracted?: Partial<DesignRequest>;
+    }>;
+}
+/**
+ * Create a requirement gatherer agent with resource-based prompts
+ */
+export declare function createRequirementGatherer(config: RequirementGathererConfig): RequirementGathererAgent;
+//# sourceMappingURL=requirement-gatherer.d.ts.map

package/dist/agents/requirement-gatherer.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"requirement-gatherer.d.ts","sourceRoot":"","sources":["../../src/agents/requirement-gatherer.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAKH,MAAM,WAAW,yBAAyB;IACxC,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,WAAW,aAAa;IAC5B,YAAY,EAAE,MAAM,CAAC;IACrB,eAAe,EAAE,MAAM,EAAE,CAAC;IAC1B,WAAW,CAAC,EAAE,MAAM,EAAE,CAAC;CACxB;AAED,MAAM,WAAW,wBAAwB;IACvC,QAAQ,EAAE;QACR,IAAI,EAAE,MAAM,CAAC;QACb,OAAO,EAAE,MAAM,CAAC;QAChB,SAAS,EAAE,0BAA0B,CAAC;KACvC,CAAC;IACF,YAAY,EAAE,MAAM,CAAC;IACrB,cAAc,EAAE,CAAC,OAAO,EAAE,MAAM,KAAK,OAAO,CAAC;QAAE,QAAQ,EAAE,MAAM,CAAC;QAAC,SAAS,CAAC,EAAE,OAAO,CAAC,aAAa,CAAC,CAAA;KAAE,CAAC,CAAC;CACxG;AAED;;GAEG;AACH,wBAAgB,yBAAyB,CAAC,MAAM,EAAE,yBAAyB,GAAG,wBAAwB,CAwCrG"}

package/dist/agents/requirement-gatherer.js

@@ -0,0 +1,57 @@
+/**
+ * Requirement Gatherer Agent
+ * Conversational assistant that extracts agent design requirements
+ */
+import * as CorePrinciples from '../../generated/resources/prompts/core-principles.js';
+import * as Questioning from '../../generated/resources/prompts/questioning-techniques.js';
+/**
+ * Create a requirement gatherer agent with resource-based prompts
+ */
+export function createRequirementGatherer(config) {
+    // Build system prompt from fragments
+    const systemPrompt = `
+You are an expert agent designer helping users define their AI agent requirements.
+
+${CorePrinciples.fragments.purposeDriven.text}
+
+${CorePrinciples.fragments.testable.text}
+
+${Questioning.text}
+
+Your job: Extract agentPurpose and successCriteria through focused questioning.
+
+Output your final understanding as JSON matching this schema:
+{
+  "agentPurpose": "string - clear, specific problem statement",
+  "successCriteria": ["string array - measurable outcomes"]
+}
+  `.trim();
+    return {
+        metadata: {
+            name: config.name,
+            version: config.version,
+            archetype: 'conversational-assistant',
+        },
+        systemPrompt,
+        /**
+         * Process a user message and extract requirements
+         */
+        async processMessage(_message) {
+            // This would integrate with an actual LLM in a real implementation
+            // For now, just demonstrate the structure
+            return {
+                response: 'Question based on core principles...',
+            };
+        },
+    };
+}
+// Example usage
+const gatherer = createRequirementGatherer({
+    name: 'requirement-gatherer',
+    version: '1.0.0',
+});
+console.log('System prompt uses resources:');
+console.log(`- Core principles fragments: ${Object.keys(CorePrinciples.fragments).length}`);
+console.log(`- Questioning techniques: ${Questioning.text.split('\n').length} lines`);
+console.log('\nAgent metadata:', gatherer.metadata);
+//# sourceMappingURL=requirement-gatherer.js.map

package/dist/agents/requirement-gatherer.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"requirement-gatherer.js","sourceRoot":"","sources":["../../src/agents/requirement-gatherer.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,cAAc,MAAM,sDAAsD,CAAC;AACvF,OAAO,KAAK,WAAW,MAAM,6DAA6D,CAAC;AAuB3F;;GAEG;AACH,MAAM,UAAU,yBAAyB,CAAC,MAAiC;IACzE,qCAAqC;IACrC,MAAM,YAAY,GAAG;;;EAGrB,cAAc,CAAC,SAAS,CAAC,aAAa,CAAC,IAAI;;EAE3C,cAAc,CAAC,SAAS,CAAC,QAAQ,CAAC,IAAI;;EAEtC,WAAW,CAAC,IAAI;;;;;;;;;GASf,CAAC,IAAI,EAAE,CAAC;IAET,OAAO;QACL,QAAQ,EAAE;YACR,IAAI,EAAE,MAAM,CAAC,IAAI;YACjB,OAAO,EAAE,MAAM,CAAC,OAAO;YACvB,SAAS,EAAE,0BAAmC;SAC/C;QACD,YAAY;QAEZ;;WAEG;QACH,KAAK,CAAC,cAAc,CAAC,QAAgB;YACnC,mEAAmE;YACnE,0CAA0C;YAE1C,OAAO;gBACL,QAAQ,EAAE,sCAAsC;aACjD,CAAC;QACJ,CAAC;KACF,CAAC;AACJ,CAAC;AAED,gBAAgB;AAChB,MAAM,QAAQ,GAAG,yBAAyB,CAAC;IACzC,IAAI,EAAE,sBAAsB;IAC5B,OAAO,EAAE,OAAO;CACjB,CAAC,CAAC;AAEH,OAAO,CAAC,GAAG,CAAC,+BAA+B,CAAC,CAAC;AAC7C,OAAO,CAAC,GAAG,CAAC,gCAAgC,MAAM,CAAC,IAAI,CAAC,cAAc,CAAC,SAAS,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC;AAC5F,OAAO,CAAC,GAAG,CAAC,6BAA6B,WAAW,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,MAAM,QAAQ,CAAC,CAAC;AACtF,OAAO,CAAC,GAAG,CAAC,mBAAmB,EAAE,QAAQ,CAAC,QAAQ,CAAC,CAAC"}

package/dist/generated/resources/prompts/core-principles.d.ts

@@ -0,0 +1,23 @@
+/**
+ * Generated TypeScript declarations - DO NOT EDIT
+ */
+
+export interface Fragment {
+  readonly header: string;
+  readonly body: string;
+  readonly text: string;
+}
+
+export const meta: {};
+
+export const text: string;
+
+export const fragments: {
+  readonly purposeDriven: Fragment;
+  readonly simpleFirst: Fragment;
+  readonly contextEfficient: Fragment;
+  readonly testable: Fragment;
+  readonly toolAppropriate: Fragment;
+};
+
+export type FragmentName = keyof typeof fragments;

package/dist/generated/resources/prompts/core-principles.js

@@ -0,0 +1,35 @@
+/**
+ * Generated from markdown file - DO NOT EDIT
+ */
+
+export const meta = {};
+
+export const text = "# Agent Design Core Principles\n\nThese are the foundational principles for designing high-quality AI agents.\n\n## Purpose Driven\n\nA well-designed agent solves a specific problem with clear success criteria.\n\n**NOT**: \"A helpful assistant\" (too vague)\n\n**YES**: \"Reviews PR diffs for security vulnerabilities, suggests fixes\"\n\n## Simple First\n\nStart with single-agent, deterministic paths.\n\n- Avoid multi-agent complexity unless clearly needed\n- Prefer simple tools over complex frameworks\n- \"Most successful implementations use simple, composable patterns\" (Anthropic)\n\n## Context Efficient\n\nUse \"smallest set of high-signal tokens\".\n\n- Prompts should provide information the LLM doesn\'t already know\n- Avoid redundancy with training data (don\'t explain common concepts)\n- Structure context hierarchically (identity → task → tools → memory)\n\n## Testable\n\nClear inputs/outputs enable validation.\n\n- Structured I/O schemas (JSON Schema format)\n- Example test cases included\n- Success measurable (accuracy, latency, cost)\n\n## Tool Appropriate\n\nRight tools for the job.\n\n- File operations? File system tools\n- Current info needed? Web search/APIs\n- Minimize tool count (each tool adds complexity)\n";
+
+export const fragments = {
+  purposeDriven: {
+    header: "## Purpose Driven",
+    body: "A well-designed agent solves a specific problem with clear success criteria.\n\n**NOT**: \"A helpful assistant\" (too vague)\n\n**YES**: \"Reviews PR diffs for security vulnerabilities, suggests fixes\"",
+    text: "## Purpose Driven\n\nA well-designed agent solves a specific problem with clear success criteria.\n\n**NOT**: \"A helpful assistant\" (too vague)\n\n**YES**: \"Reviews PR diffs for security vulnerabilities, suggests fixes\""
+  },
+  simpleFirst: {
+    header: "## Simple First",
+    body: "Start with single-agent, deterministic paths.\n\n- Avoid multi-agent complexity unless clearly needed\n- Prefer simple tools over complex frameworks\n- \"Most successful implementations use simple, composable patterns\" (Anthropic)",
+    text: "## Simple First\n\nStart with single-agent, deterministic paths.\n\n- Avoid multi-agent complexity unless clearly needed\n- Prefer simple tools over complex frameworks\n- \"Most successful implementations use simple, composable patterns\" (Anthropic)"
+  },
+  contextEfficient: {
+    header: "## Context Efficient",
+    body: "Use \"smallest set of high-signal tokens\".\n\n- Prompts should provide information the LLM doesn\'t already know\n- Avoid redundancy with training data (don\'t explain common concepts)\n- Structure context hierarchically (identity → task → tools → memory)",
+    text: "## Context Efficient\n\nUse \"smallest set of high-signal tokens\".\n\n- Prompts should provide information the LLM doesn\'t already know\n- Avoid redundancy with training data (don\'t explain common concepts)\n- Structure context hierarchically (identity → task → tools → memory)"
+  },
+  testable: {
+    header: "## Testable",
+    body: "Clear inputs/outputs enable validation.\n\n- Structured I/O schemas (JSON Schema format)\n- Example test cases included\n- Success measurable (accuracy, latency, cost)",
+    text: "## Testable\n\nClear inputs/outputs enable validation.\n\n- Structured I/O schemas (JSON Schema format)\n- Example test cases included\n- Success measurable (accuracy, latency, cost)"
+  },
+  toolAppropriate: {
+    header: "## Tool Appropriate",
+    body: "Right tools for the job.\n\n- File operations? File system tools\n- Current info needed? Web search/APIs\n- Minimize tool count (each tool adds complexity)",
+    text: "## Tool Appropriate\n\nRight tools for the job.\n\n- File operations? File system tools\n- Current info needed? Web search/APIs\n- Minimize tool count (each tool adds complexity)"
+  }
+};

package/dist/generated/resources/prompts/llm-selection.d.ts

@@ -0,0 +1,22 @@
+/**
+ * Generated TypeScript declarations - DO NOT EDIT
+ */
+
+export interface Fragment {
+  readonly header: string;
+  readonly body: string;
+  readonly text: string;
+}
+
+export const meta: {};
+
+export const text: string;
+
+export const fragments: {
+  readonly claudeModels: Fragment;
+  readonly openaiModels: Fragment;
+  readonly selectionCriteria: Fragment;
+  readonly commonMistakes: Fragment;
+};
+
+export type FragmentName = keyof typeof fragments;

package/dist/generated/resources/prompts/llm-selection.js

@@ -0,0 +1,30 @@
+/**
+ * Generated from markdown file - DO NOT EDIT
+ */
+
+export const meta = {};
+
+export const text = "# LLM Selection Guidance\n\nGuidelines for selecting the right LLM based on agent requirements.\n\n## Claude Models\n\n**Claude 3.5 Sonnet:**\n- **Best for:** Complex reasoning, coding, analysis\n- **Strengths:** Long context (200K), tool use, function calling\n- **Cost:** Mid-range ($3/$15 per 1M tokens)\n- **Latency:** ~2-5 seconds\n\n**Claude 3 Haiku:**\n- **Best for:** Fast, simple tasks, high-volume\n- **Strengths:** Speed, cost-effective\n- **Cost:** Low ($0.25/$1.25 per 1M tokens)\n- **Latency:** ~500ms-1s\n\n## OpenAI Models\n\n**GPT-4 Turbo:**\n- **Best for:** Complex reasoning, multimodal\n- **Strengths:** Vision, JSON mode, function calling\n- **Cost:** Mid-range ($10/$30 per 1M tokens)\n- **Latency:** ~2-4 seconds\n\n**GPT-3.5 Turbo:**\n- **Best for:** Simple tasks, cost-sensitive\n- **Strengths:** Fast, cheap, widely deployed\n- **Cost:** Very low ($0.50/$1.50 per 1M tokens)\n- **Latency:** ~500ms-1s\n\n## Selection Criteria\n\n**Choose by latency:**\n- Real-time (<1s): Haiku, GPT-3.5\n- Interactive (<5s): Sonnet, GPT-4\n- Batch (>5s acceptable): Any model\n\n**Choose by complexity:**\n- Simple classification: Haiku, GPT-3.5\n- Reasoning/analysis: Sonnet, GPT-4\n- Complex multi-step: Sonnet, GPT-4 Turbo\n\n**Choose by cost:**\n- High volume: Haiku, GPT-3.5\n- Medium volume: Sonnet\n- Low volume/critical: GPT-4, Sonnet\n\n**Choose by features:**\n- Vision needed: GPT-4 Turbo (multimodal)\n- Long context: Claude (200K vs 128K)\n- Structured output: GPT-4 (native JSON mode)\n- Tool use: Claude (best-in-class)\n\n## Common Mistakes\n\n**Over-engineering:**\n- Using GPT-4 for simple classification (use Haiku/GPT-3.5)\n- Using latest model when older works fine\n\n**Under-engineering:**\n- Using GPT-3.5 for complex reasoning (upgrade to Sonnet/GPT-4)\n- Skipping evaluation (always benchmark)\n\n**Cost blindness:**\n- Not considering volume × price\n- Ignoring prompt optimization opportunities\n";
+
+export const fragments = {
+  claudeModels: {
+    header: "## Claude Models",
+    body: "**Claude 3.5 Sonnet:**\n- **Best for:** Complex reasoning, coding, analysis\n- **Strengths:** Long context (200K), tool use, function calling\n- **Cost:** Mid-range ($3/$15 per 1M tokens)\n- **Latency:** ~2-5 seconds\n\n**Claude 3 Haiku:**\n- **Best for:** Fast, simple tasks, high-volume\n- **Strengths:** Speed, cost-effective\n- **Cost:** Low ($0.25/$1.25 per 1M tokens)\n- **Latency:** ~500ms-1s",
+    text: "## Claude Models\n\n**Claude 3.5 Sonnet:**\n- **Best for:** Complex reasoning, coding, analysis\n- **Strengths:** Long context (200K), tool use, function calling\n- **Cost:** Mid-range ($3/$15 per 1M tokens)\n- **Latency:** ~2-5 seconds\n\n**Claude 3 Haiku:**\n- **Best for:** Fast, simple tasks, high-volume\n- **Strengths:** Speed, cost-effective\n- **Cost:** Low ($0.25/$1.25 per 1M tokens)\n- **Latency:** ~500ms-1s"
+  },
+  openaiModels: {
+    header: "## OpenAI Models",
+    body: "**GPT-4 Turbo:**\n- **Best for:** Complex reasoning, multimodal\n- **Strengths:** Vision, JSON mode, function calling\n- **Cost:** Mid-range ($10/$30 per 1M tokens)\n- **Latency:** ~2-4 seconds\n\n**GPT-3.5 Turbo:**\n- **Best for:** Simple tasks, cost-sensitive\n- **Strengths:** Fast, cheap, widely deployed\n- **Cost:** Very low ($0.50/$1.50 per 1M tokens)\n- **Latency:** ~500ms-1s",
+    text: "## OpenAI Models\n\n**GPT-4 Turbo:**\n- **Best for:** Complex reasoning, multimodal\n- **Strengths:** Vision, JSON mode, function calling\n- **Cost:** Mid-range ($10/$30 per 1M tokens)\n- **Latency:** ~2-4 seconds\n\n**GPT-3.5 Turbo:**\n- **Best for:** Simple tasks, cost-sensitive\n- **Strengths:** Fast, cheap, widely deployed\n- **Cost:** Very low ($0.50/$1.50 per 1M tokens)\n- **Latency:** ~500ms-1s"
+  },
+  selectionCriteria: {
+    header: "## Selection Criteria",
+    body: "**Choose by latency:**\n- Real-time (<1s): Haiku, GPT-3.5\n- Interactive (<5s): Sonnet, GPT-4\n- Batch (>5s acceptable): Any model\n\n**Choose by complexity:**\n- Simple classification: Haiku, GPT-3.5\n- Reasoning/analysis: Sonnet, GPT-4\n- Complex multi-step: Sonnet, GPT-4 Turbo\n\n**Choose by cost:**\n- High volume: Haiku, GPT-3.5\n- Medium volume: Sonnet\n- Low volume/critical: GPT-4, Sonnet\n\n**Choose by features:**\n- Vision needed: GPT-4 Turbo (multimodal)\n- Long context: Claude (200K vs 128K)\n- Structured output: GPT-4 (native JSON mode)\n- Tool use: Claude (best-in-class)",
+    text: "## Selection Criteria\n\n**Choose by latency:**\n- Real-time (<1s): Haiku, GPT-3.5\n- Interactive (<5s): Sonnet, GPT-4\n- Batch (>5s acceptable): Any model\n\n**Choose by complexity:**\n- Simple classification: Haiku, GPT-3.5\n- Reasoning/analysis: Sonnet, GPT-4\n- Complex multi-step: Sonnet, GPT-4 Turbo\n\n**Choose by cost:**\n- High volume: Haiku, GPT-3.5\n- Medium volume: Sonnet\n- Low volume/critical: GPT-4, Sonnet\n\n**Choose by features:**\n- Vision needed: GPT-4 Turbo (multimodal)\n- Long context: Claude (200K vs 128K)\n- Structured output: GPT-4 (native JSON mode)\n- Tool use: Claude (best-in-class)"
+  },
+  commonMistakes: {
+    header: "## Common Mistakes",
+    body: "**Over-engineering:**\n- Using GPT-4 for simple classification (use Haiku/GPT-3.5)\n- Using latest model when older works fine\n\n**Under-engineering:**\n- Using GPT-3.5 for complex reasoning (upgrade to Sonnet/GPT-4)\n- Skipping evaluation (always benchmark)\n\n**Cost blindness:**\n- Not considering volume × price\n- Ignoring prompt optimization opportunities",
+    text: "## Common Mistakes\n\n**Over-engineering:**\n- Using GPT-4 for simple classification (use Haiku/GPT-3.5)\n- Using latest model when older works fine\n\n**Under-engineering:**\n- Using GPT-3.5 for complex reasoning (upgrade to Sonnet/GPT-4)\n- Skipping evaluation (always benchmark)\n\n**Cost blindness:**\n- Not considering volume × price\n- Ignoring prompt optimization opportunities"
+  }
+};

package/dist/generated/resources/prompts/pattern-recognition.d.ts

@@ -0,0 +1,23 @@
+/**
+ * Generated TypeScript declarations - DO NOT EDIT
+ */
+
+export interface Fragment {
+  readonly header: string;
+  readonly body: string;
+  readonly text: string;
+}
+
+export const meta: {};
+
+export const text: string;
+
+export const fragments: {
+  readonly pureFunctionToolPattern: Fragment;
+  readonly llmAnalyzerPattern: Fragment;
+  readonly conversationalAssistantPattern: Fragment;
+  readonly agenticWorkflowPattern: Fragment;
+  readonly redFlags: Fragment;
+};
+
+export type FragmentName = keyof typeof fragments;

package/dist/generated/resources/prompts/pattern-recognition.js

@@ -0,0 +1,35 @@
+/**
+ * Generated from markdown file - DO NOT EDIT
+ */
+
+export const meta = {};
+
+export const text = "# Agent Pattern Recognition\n\nGuidance for identifying the right agent archetype based on requirements.\n\n## Pure Function Tool Pattern\n\n**Recognizable by:**\n- Stateless transformation (same input → same output)\n- No conversation needed\n- Clear input/output schema\n- No external dependencies or context\n\n**Examples:**\n- JSON validator\n- Text formatter\n- Data transformer\n- Schema converter\n\n**Best for:** Deterministic operations, utilities, data processing\n\n## LLM Analyzer Pattern\n\n**Recognizable by:**\n- Analysis of unstructured input\n- Produces structured output\n- Single-turn operation\n- No back-and-forth conversation\n\n**Examples:**\n- Sentiment analyzer\n- Code reviewer\n- Document summarizer\n- Requirements extractor\n\n**Best for:** Classification, analysis, extraction, evaluation\n\n## Conversational Assistant Pattern\n\n**Recognizable by:**\n- Multi-turn dialogue\n- Context accumulation across turns\n- Clarifying questions\n- Adaptive responses based on history\n\n**Examples:**\n- Requirements gathering chatbot\n- Technical advisor\n- Interactive troubleshooter\n- Teaching assistant\n\n**Best for:** Exploration, education, consultation, complex requirements\n\n## Agentic Workflow Pattern\n\n**Recognizable by:**\n- Multiple steps/decisions\n- Tool use required\n- External system integration\n- State management across actions\n\n**Examples:**\n- Code generator with file operations\n- Research assistant with web search\n- Data pipeline orchestrator\n- Multi-tool problem solver\n\n**Best for:** Complex automation, integration, multi-step processes\n\n## Red Flags\n\n**When to NOT use agents:**\n- Simple rule-based logic (use if/else)\n- Known algorithm exists (use traditional code)\n- Real-time requirements (<100ms)\n- Deterministic correctness required (use formal verification)\n";
+
+export const fragments = {
+  pureFunctionToolPattern: {
+    header: "## Pure Function Tool Pattern",
+    body: "**Recognizable by:**\n- Stateless transformation (same input → same output)\n- No conversation needed\n- Clear input/output schema\n- No external dependencies or context\n\n**Examples:**\n- JSON validator\n- Text formatter\n- Data transformer\n- Schema converter\n\n**Best for:** Deterministic operations, utilities, data processing",
+    text: "## Pure Function Tool Pattern\n\n**Recognizable by:**\n- Stateless transformation (same input → same output)\n- No conversation needed\n- Clear input/output schema\n- No external dependencies or context\n\n**Examples:**\n- JSON validator\n- Text formatter\n- Data transformer\n- Schema converter\n\n**Best for:** Deterministic operations, utilities, data processing"
+  },
+  llmAnalyzerPattern: {
+    header: "## LLM Analyzer Pattern",
+    body: "**Recognizable by:**\n- Analysis of unstructured input\n- Produces structured output\n- Single-turn operation\n- No back-and-forth conversation\n\n**Examples:**\n- Sentiment analyzer\n- Code reviewer\n- Document summarizer\n- Requirements extractor\n\n**Best for:** Classification, analysis, extraction, evaluation",
+    text: "## LLM Analyzer Pattern\n\n**Recognizable by:**\n- Analysis of unstructured input\n- Produces structured output\n- Single-turn operation\n- No back-and-forth conversation\n\n**Examples:**\n- Sentiment analyzer\n- Code reviewer\n- Document summarizer\n- Requirements extractor\n\n**Best for:** Classification, analysis, extraction, evaluation"
+  },
+  conversationalAssistantPattern: {
+    header: "## Conversational Assistant Pattern",
+    body: "**Recognizable by:**\n- Multi-turn dialogue\n- Context accumulation across turns\n- Clarifying questions\n- Adaptive responses based on history\n\n**Examples:**\n- Requirements gathering chatbot\n- Technical advisor\n- Interactive troubleshooter\n- Teaching assistant\n\n**Best for:** Exploration, education, consultation, complex requirements",
+    text: "## Conversational Assistant Pattern\n\n**Recognizable by:**\n- Multi-turn dialogue\n- Context accumulation across turns\n- Clarifying questions\n- Adaptive responses based on history\n\n**Examples:**\n- Requirements gathering chatbot\n- Technical advisor\n- Interactive troubleshooter\n- Teaching assistant\n\n**Best for:** Exploration, education, consultation, complex requirements"
+  },
+  agenticWorkflowPattern: {
+    header: "## Agentic Workflow Pattern",
+    body: "**Recognizable by:**\n- Multiple steps/decisions\n- Tool use required\n- External system integration\n- State management across actions\n\n**Examples:**\n- Code generator with file operations\n- Research assistant with web search\n- Data pipeline orchestrator\n- Multi-tool problem solver\n\n**Best for:** Complex automation, integration, multi-step processes",
+    text: "## Agentic Workflow Pattern\n\n**Recognizable by:**\n- Multiple steps/decisions\n- Tool use required\n- External system integration\n- State management across actions\n\n**Examples:**\n- Code generator with file operations\n- Research assistant with web search\n- Data pipeline orchestrator\n- Multi-tool problem solver\n\n**Best for:** Complex automation, integration, multi-step processes"
+  },
+  redFlags: {
+    header: "## Red Flags",
+    body: "**When to NOT use agents:**\n- Simple rule-based logic (use if/else)\n- Known algorithm exists (use traditional code)\n- Real-time requirements (<100ms)\n- Deterministic correctness required (use formal verification)",
+    text: "## Red Flags\n\n**When to NOT use agents:**\n- Simple rule-based logic (use if/else)\n- Known algorithm exists (use traditional code)\n- Real-time requirements (<100ms)\n- Deterministic correctness required (use formal verification)"
+  }
+};

package/dist/generated/resources/prompts/questioning-techniques.d.ts

@@ -0,0 +1,22 @@
+/**
+ * Generated TypeScript declarations - DO NOT EDIT
+ */
+
+export interface Fragment {
+  readonly header: string;
+  readonly body: string;
+  readonly text: string;
+}
+
+export const meta: {};
+
+export const text: string;
+
+export const fragments: {
+  readonly askOneAtATime: Fragment;
+  readonly pushForSpecificity: Fragment;
+  readonly extractConcreteRequirements: Fragment;
+  readonly avoidOpenEndedQuestions: Fragment;
+};
+
+export type FragmentName = keyof typeof fragments;

package/dist/generated/resources/prompts/questioning-techniques.js

@@ -0,0 +1,30 @@
+/**
+ * Generated from markdown file - DO NOT EDIT
+ */
+
+export const meta = {};
+
+export const text = "# Questioning Techniques for Requirements Gathering\n\nEffective questioning strategies for eliciting agent requirements.\n\n## Ask One at a Time\n\nDon\'t overwhelm users with multiple questions at once. Focus on one concern at a time.\n\n**Good**: \"What problem are you solving?\"\n\n**Bad**: \"Tell me about your agent, what it does, who uses it, and what tools it needs\"\n\n## Push for Specificity\n\nWhen answers are vague, follow up with targeted questions.\n\n**Vague answer**: \"It helps with coding\"\n**Follow-up**: \"What specific coding tasks? Code review? Generation? Refactoring?\"\n\n## Extract Concrete Requirements\n\nAlways aim to extract:\n- **Problem statement**: Who, what, why\n- **Success criteria**: Measurable outcomes\n- **Constraints**: Performance, cost, integration needs\n\n## Avoid Open-Ended Questions\n\n**Bad**: \"Tell me about your use case\"\n**Good**: \"What problem are you solving?\" (specific, actionable)\n";
+
+export const fragments = {
+  askOneAtATime: {
+    header: "## Ask One at a Time",
+    body: "Don\'t overwhelm users with multiple questions at once. Focus on one concern at a time.\n\n**Good**: \"What problem are you solving?\"\n\n**Bad**: \"Tell me about your agent, what it does, who uses it, and what tools it needs\"",
+    text: "## Ask One at a Time\n\nDon\'t overwhelm users with multiple questions at once. Focus on one concern at a time.\n\n**Good**: \"What problem are you solving?\"\n\n**Bad**: \"Tell me about your agent, what it does, who uses it, and what tools it needs\""
+  },
+  pushForSpecificity: {
+    header: "## Push for Specificity",
+    body: "When answers are vague, follow up with targeted questions.\n\n**Vague answer**: \"It helps with coding\"\n**Follow-up**: \"What specific coding tasks? Code review? Generation? Refactoring?\"",
+    text: "## Push for Specificity\n\nWhen answers are vague, follow up with targeted questions.\n\n**Vague answer**: \"It helps with coding\"\n**Follow-up**: \"What specific coding tasks? Code review? Generation? Refactoring?\""
+  },
+  extractConcreteRequirements: {
+    header: "## Extract Concrete Requirements",
+    body: "Always aim to extract:\n- **Problem statement**: Who, what, why\n- **Success criteria**: Measurable outcomes\n- **Constraints**: Performance, cost, integration needs",
+    text: "## Extract Concrete Requirements\n\nAlways aim to extract:\n- **Problem statement**: Who, what, why\n- **Success criteria**: Measurable outcomes\n- **Constraints**: Performance, cost, integration needs"
+  },
+  avoidOpenEndedQuestions: {
+    header: "## Avoid Open-Ended Questions",
+    body: "**Bad**: \"Tell me about your use case\"\n**Good**: \"What problem are you solving?\" (specific, actionable)",
+    text: "## Avoid Open-Ended Questions\n\n**Bad**: \"Tell me about your use case\"\n**Good**: \"What problem are you solving?\" (specific, actionable)"
+  }
+};

package/dist/generated/resources/skills/SKILL.d.ts

@@ -0,0 +1,35 @@
+/**
+ * Generated TypeScript declarations - DO NOT EDIT
+ */
+
+export interface Fragment {
+  readonly header: string;
+  readonly body: string;
+  readonly text: string;
+}
+
+export const meta: {
+  readonly name: string;
+  readonly description: string;
+};
+
+export const text: string;
+
+export const fragments: {
+  readonly purposeForUsersNotContributors: Fragment;
+  readonly whatIsVat: Fragment;
+  readonly whenToUseVat: Fragment;
+  readonly creatingAgentsWithAgentGenerator: Fragment;
+  readonly understandingAgentPatterns: Fragment;
+  readonly vatCliWorkflows: Fragment;
+  readonly resultEnvelopesAndErrorHandling: Fragment;
+  readonly orchestrationPatterns: Fragment;
+  readonly testingAgents: Fragment;
+  readonly bestPractices: Fragment;
+  readonly nextSteps: Fragment;
+  readonly documentationIndex: Fragment;
+  readonly successCriteria: Fragment;
+  readonly gettingHelp: Fragment;
+};
+
+export type FragmentName = keyof typeof fragments;