#!/usr/bin/env node
/**
* Intent analysis orchestrator
*
* Coordinates AI-powered intent analysis using modular components:
* - Anthropic API client for skill scoring
* - Cache manager for result persistence
* - Keyword matcher for fallback
* - Intent scorer for categorization
*/
import { createHash } from "crypto";
import { SHORT_PROMPT_WORD_THRESHOLD } from "./constants.js";
import { readCache, writeCache } from "./cache-manager.js";
import { callAnthropicAPI } from "./anthropic-client.js";
import { matchSkillsByKeywords } from "./keyword-matcher.js";
import {
categorizeSkills,
formatDebugOutput,
buildAnalysisResult
} from "./intent-scorer.js";
import type { AnalysisResult, SkillRule } from "./types.js";
// Re-export types for backward compatibility
export type {
SkillConfidence,
IntentAnalysis,
AnalysisResult
} from "./types.js";
/**
* Analyzes user intent using AI to determine relevant skills
*
* Uses Claude API to analyze the user's prompt and assign confidence
* scores to each skill. Model defaults to Haiku 4.5 but can be configured
* via CLAUDE_SKILLS_MODEL env var. Falls back to keyword matching for short
* prompts (<10 words) or if AI analysis fails. Results are cached for 1 hour.
*
* @param prompt - The user's input prompt to analyze
* @param availableSkills - Record of skill configurations from skill-rules.json
* @returns Promise resolving to required/suggested skill lists with optional scores
*
* @example
* const result = await analyzeIntent("Fix Java adapter", skillRules);
* // Returns: { required: ['adapter-development'], suggested: [], fromCache: false }
*/
export async function analyzeIntent(
prompt: string,
availableSkills: Record<string, SkillRule>
): Promise<AnalysisResult> {
// Skip AI analysis for short prompts (saves API calls)
const wordCount = prompt.trim().split(/\s+/).length;
if (wordCount <= SHORT_PROMPT_WORD_THRESHOLD) {
return matchSkillsByKeywords(prompt, availableSkills);
}
// Check cache first - include skills hash to invalidate when definitions change
const skillsHash = createHash("md5")
.update(JSON.stringify(availableSkills))
.digest("hex")
.substring(0, 8);
const cacheKey = createHash("md5")
.update(prompt + skillsHash)
.digest("hex");
const cached = readCache(cacheKey);
if (cached) {
return { ...cached, fromCache: true };
}
// Call Anthropic API
try {
const analysis = await callAnthropicAPI(prompt, availableSkills);
// Debug logging
if (process.env.AIDB_SKILL_DEBUG === "true") {
formatDebugOutput(analysis);
}
// Categorize by confidence thresholds
const categorized = categorizeSkills(analysis);
// Build result with optional debug scores
const result = buildAnalysisResult(
categorized,
analysis,
process.env.AIDB_SKILL_DEBUG === "true"
);
writeCache(cacheKey, {
required: result.required,
suggested: result.suggested
});
return result;
} catch (error) {
console.warn(
"Intent analysis failed, falling back to keyword matching:",
error
);
return matchSkillsByKeywords(prompt, availableSkills);
}
}
