load_prompt
Retrieve prompt templates for ADR generation and analysis on-demand, reducing token usage by loading only when needed.
Instructions
Load a specific prompt or prompt section on-demand. Part of CE-MCP lazy loading system that reduces token usage by ~96% by loading prompts only when needed. Use this to retrieve prompt templates for ADR generation, analysis, deployment, and other operations.
Input Schema
TableJSON Schema
| Name | Required | Description | Default |
|---|---|---|---|
| promptName | Yes | Name of the prompt to load (e.g., "adr-suggestion", "deployment-analysis", "environment-analysis", "research-question", "rule-generation", "analysis", "security") | |
| section | No | Specific section within the prompt to load. If not provided, loads the entire prompt. Available sections depend on the prompt. | |
| estimateOnly | No | If true, returns only token estimate without loading the full prompt content |
Implementation Reference
- src/tools/tool-catalog.ts:1362-1403 (registration)Registration of the 'load_prompt' MCP tool in the central TOOL_CATALOG, including metadata, description, input schema, and CE-MCP support.
TOOL_CATALOG.set('load_prompt', { name: 'load_prompt', shortDescription: 'Load prompts on-demand (CE-MCP)', fullDescription: 'Loads prompts on-demand instead of eagerly loading all prompts at startup. Part of CE-MCP lazy loading system that reduces token usage by ~96%.', category: 'utility', complexity: 'simple', tokenCost: { min: 100, max: 500 }, hasCEMCPDirective: true, relatedTools: ['search_tools', 'analyze_project_ecosystem'], keywords: ['prompt', 'load', 'lazy', 'ce-mcp', 'token', 'optimization'], requiresAI: false, inputSchema: { type: 'object', properties: { promptName: { type: 'string', description: 'Name of the prompt to load', enum: [ 'adr-suggestion', 'deployment-analysis', 'environment-analysis', 'research-question', 'rule-generation', 'analysis', 'research-integration', 'validated-pattern', 'security', ], }, section: { type: 'string', description: 'Specific section within the prompt to load', }, estimateOnly: { type: 'boolean', description: 'Return only token estimate without loading content', }, }, required: ['promptName'], }, }); - src/prompts/prompt-catalog.ts:167-209 (handler)Primary handler logic for loading prompts on-demand with caching, section extraction, and metadata validation using PROMPT_CATALOG. This implements the core functionality of the load_prompt tool.
async loadPrompt(promptName: string, section?: string): Promise<string> { const metadata = PROMPT_CATALOG[promptName]; if (!metadata) { throw new Error(`Unknown prompt: ${promptName}`); } // Check cache const cacheKey = promptName; const cached = this.cache.get(cacheKey); if (cached && Date.now() < cached.expiry) { if (section) { const sectionContent = cached.sections.get(section); if (sectionContent) { return sectionContent; } } return cached.content; } // Load prompt dynamically const content = await this.loadPromptFile(metadata.file); // Parse sections if needed const sections = this.parseSections(content, metadata.sections); // Cache the result this.cache.set(cacheKey, { content, sections, expiry: Date.now() + this.cacheTTL * 1000, }); if (section) { const sectionContent = sections.get(section); if (sectionContent) { return sectionContent; } throw new Error(`Section '${section}' not found in prompt '${promptName}'`); } return content; } - src/prompts/prompt-catalog.ts:18-121 (helper)PROMPT_CATALOG constant providing metadata for all available prompts, including file paths, token estimates, categories, sections, and load-on-demand flags. Essential helper for prompt loading.
export const PROMPT_CATALOG: PromptCatalog = { // ADR-related prompts 'adr-suggestion': { file: 'adr-suggestion-prompts.ts', tokens: 1830, category: 'adr', sections: [ 'implicit_decisions', 'tech_debt', 'security_decisions', 'cross_cutting', 'recommendation_template', ], loadOnDemand: true, }, // Deployment analysis prompts 'deployment-analysis': { file: 'deployment-analysis-prompts.ts', tokens: 3150, category: 'deployment', sections: [ 'readiness_check', 'validation_criteria', 'rollback_plan', 'infrastructure_review', 'security_scan', ], loadOnDemand: true, }, // Environment analysis prompts 'environment-analysis': { file: 'environment-analysis-prompts.ts', tokens: 3050, category: 'analysis', sections: ['dependency_scan', 'config_validation', 'compliance_check', 'resource_assessment'], loadOnDemand: true, }, // Research question prompts 'research-question': { file: 'research-question-prompts.ts', tokens: 3120, category: 'research', sections: ['question_generation', 'research_plan', 'source_evaluation', 'synthesis_template'], loadOnDemand: true, }, // Rule generation prompts 'rule-generation': { file: 'rule-generation-prompts.ts', tokens: 2850, category: 'rules', sections: ['rule_template', 'validation_rules', 'enforcement_policy', 'exception_handling'], loadOnDemand: true, }, // General analysis prompts analysis: { file: 'analysis-prompts.ts', tokens: 2310, category: 'analysis', sections: ['project_analysis', 'code_review', 'architecture_assessment', 'quality_metrics'], loadOnDemand: true, }, // Research integration prompts 'research-integration': { file: 'research-integration-prompts.ts', tokens: 1785, category: 'research', sections: ['integration_strategy', 'synthesis_plan', 'recommendation_format'], loadOnDemand: true, }, // Validated pattern prompts 'validated-pattern': { file: 'validated-pattern-prompts.ts', tokens: 1565, category: 'deployment', sections: ['pattern_detection', 'validation_criteria', 'deployment_guidance'], loadOnDemand: true, }, // Security prompts security: { file: 'security-prompts.ts', tokens: 1270, category: 'security', sections: ['vulnerability_scan', 'masking_rules', 'compliance_check'], loadOnDemand: true, }, // Main index (orchestration) index: { file: 'index.ts', tokens: 3700, category: 'analysis', sections: ['orchestration', 'execution_flow', 'error_handling'], dependencies: ['adr-suggestion', 'deployment-analysis', 'analysis'], loadOnDemand: false, // Core orchestration loaded at startup }, }; - src/tools/tool-catalog.ts:1374-1402 (schema)Input schema definition for the load_prompt tool, specifying parameters like promptName (with enum), section, and estimateOnly.
inputSchema: { type: 'object', properties: { promptName: { type: 'string', description: 'Name of the prompt to load', enum: [ 'adr-suggestion', 'deployment-analysis', 'environment-analysis', 'research-question', 'rule-generation', 'analysis', 'research-integration', 'validated-pattern', 'security', ], }, section: { type: 'string', description: 'Specific section within the prompt to load', }, estimateOnly: { type: 'boolean', description: 'Return only token estimate without loading content', }, }, required: ['promptName'], }, - Internal sandbox operation handler opLoadPrompt used by CE-MCP directives for lazy prompt loading within orchestrated workflows.
private async opLoadPrompt( args: Record<string, unknown> | undefined, context: SandboxContext ): Promise<string> { const promptName = args?.['name'] as string; const section = args?.['section'] as string | undefined; if (!promptName) { throw new Error('loadPrompt requires "name" argument'); } // Check prompt cache with LRU tracking const cacheKey = `prompt:${promptName}:${section || 'full'}`; const cached = this.promptCache.get(cacheKey); if (cached && Date.now() < cached.expiry) { cached.lastAccess = Date.now(); this.cacheHits++; return cached.result; } this.cacheMisses++; if (cached) { this.promptCache.delete(cacheKey); } // Load prompt from file system // In full implementation, would use prompt catalog const promptPath = join(context.projectPath, 'src', 'prompts', `${promptName}.ts`); try { const content = await readFile(promptPath, 'utf-8'); // Evict old entries if cache is full this.evictOldestEntries(this.promptCache, MAX_PROMPT_CACHE_SIZE); // Cache the result with LRU tracking this.promptCache.set(cacheKey, { result: content, expiry: Date.now() + this.config.prompts.cacheTTL * 1000, lastAccess: Date.now(), }); return content; } catch { // Return placeholder if prompt not found return `[Prompt: ${promptName}${section ? `:${section}` : ''}]`; } }