import { LRUCache } from 'lru-cache';
// Simplified cache config (v11 compatible)
const promptCache = new LRUCache<string, string>({
max: 100,
ttl: 60 * 60 * 1000, // 1 hour TTL
allowStale: false,
updateAgeOnGet: true
});
export const systemPrompt = () => {
const now = new Date();
const hour = now.getHours();
const researchPhase = () => {
return hour < 6 ? 'Night: Batch processing' :
hour < 12 ? 'Morning: Discovery' :
hour < 18 ? 'Afternoon: Validation' :
'Evening: Synthesis';
};
const instructions = [
"PRISMA systematic reviews w/ source hierarchy: peer-reviewed > preprints > reports > media",
"CASP quality checks + thematic/meta-analysis",
"Contradiction documentation w/ evidence weighting",
"Authoritative source prioritization + technical depth",
"Doctorate-level analysis w/ innovation forecasting",
"Structured explanations + citation rigor",
"Glossary appendix + emerging tech exploration",
"Speculative content labeling + iterative integration",
"Markdown formatting + self-correcting workflow"
];
// Web Research-Based Improvement: Add response format template
const responseFormat = [
"**Response Structure:**",
"1. Executive Summary (3-5 bullet points)",
"2. Key Findings (categorized by relevance)",
"3. Methodology Description",
"4. Limitations Disclosure",
"5. Future Research Directions",
"6. Glossary of Technical Terms"
].join("\n");
// Add time-based research parameters
const timeContext = hour < 12
? "Morning Research Protocol: Aggressive discovery phase"
: "Afternoon Research Protocol: Validation & synthesis focus";
return `## Research Agent Configuration
**Temporal Context:** ${timeContext}
**System Time:** ${now.toLocaleDateString('en-US', {
weekday: 'long',
year: 'numeric',
month: 'long',
day: 'numeric',
hour: '2-digit',
minute: '2-digit',
timeZoneName: 'short'
})}
**Research Protocol:** v2.3.1
**Model Capabilities:** ${process.env.GEMINI_MODEL || "gemini-2.5-flash"}
**Research Phase:** ${researchPhase()}
**Cognitive Load Profile:** ${hour < 12 ? 'High creativity' : 'High accuracy'} mode
${responseFormat}
## Operational Parameters
${instructions.map((i, idx) => `${idx + 1}. ${i}`).join("\n")}`;
};
export const serpQueryPromptTemplate = `## Search Strategy Configuration
**Query Type Distribution:**
- Informational: 40%
- Comparative: 25%
- Technical: 20%
- Exploratory: 15%
**Response Format Requirements:** (Return ONLY JSON matching this schema. No extra text.)
\`\`\`json
{
"queries": {
"type": "array",
"items": {
"type": "object",
"properties": {
"query": {"type": "string", "maxLength": 60},
"type": {"type": "string", "enum": ["informational", "comparative", "technical", "exploratory"]},
"expectedSources": {"type": "array", "items": {"type": "string"}}
},
"required": ["query", "type"]
},
"minItems": {{numQueries}},
"maxItems": {{numQueries}}
}
}
\`\`\``;
export const learningPromptTemplate = `## Content Analysis Protocol
**Schema Version:** 2.1.0
**Error Handling:**
- 5001: Invalid source credibility score
- 5002: Missing fact verification
- 5003: Technical term mismatch
{
"$schema": "http://json-schema.org/draft-07/schema#",
"errorCodes": {
"5001": "Source score must be 1-5",
"5002": "Minimum 3 verifications required",
"5003": "Undefined technical terms present"
},
"type": "object",
"properties": {
"learnings": {
"type": "array",
"items": {
"type": "string",
"maxLength": 280,
"pattern": "^[A-Z][^.]{10,255}\\."
}
}
},
"required": ["learnings"],
"additionalProperties": false
}
Return ONLY JSON that matches this schema. No extra text.`;
export const feedbackPromptTemplate = `## Query Refinement Matrix
**Quality Metrics:**
- Specificity Score (1-5)
- Research Depth Potential
- Answerability Estimate
**Response Schema:** (Return ONLY JSON matching this schema. No extra text.)
\`\`\`json
{
"followUpQuestions": {
"type": "array",
"items": {
"type": "object",
"properties": {
"question": {"type": "string"},
"purpose": {"type": "string", "enum": ["scope", "context", "precision"]},
"criticality": {"type": "number", "minimum": 1, "maximum": 3}
}
}
}
}
\`\`\``;
// --- REVISED generateGeminiPrompt FUNCTION ---
export const generateGeminiPrompt = ({
query, // Now used
researchGoal, // Now used
learnings // Now used
}: {
query: string,
researchGoal: string,
learnings: string[]
}): string => {
// Implement actual logic using all parameters
return `Research Goal: ${researchGoal}
Query: ${query}
Learnings: ${learnings.join(', ')}`;
};
// --- END REVISED generateGeminiPrompt FUNCTION ---
// Keep only what's actually used
export const clearPromptCache = () => promptCache.clear();
// Add validation layer
const validatePromptConsistency = () => {
const templates = [systemPrompt, serpQueryPromptTemplate, learningPromptTemplate];
const versions = templates.map(t => {
const content = typeof t === 'function' ? t() : t;
return content.match(/Schema Version: (\d+\.\d+\.\d+)/)?.[1];
});
if (new Set(versions).size > 1) {
throw new Error(`Prompt version mismatch: ${versions.join(' vs ')}`);
}
};
validatePromptConsistency(); // Run at module load
