[#15] AI가 deterministic evidence 기반 prediction과 action을 제안한다

[opficdev]

🔗 연관된 이슈

• closed AI가 deterministic evidence 기반 prediction과 action을 제안한다 #15

🎯 의도

deterministic possibility 결과를 기반으로 AI prediction을 생성하기 위한 provider 독립 core 구성

📝 작업 내용

📌 요약

• AI prediction 모델 구성
• deterministic evidence payload 구성
• AI prediction 대상 선택 기준 구성
• AI prediction 응답 검증 구성
• AI prediction prompt 구성
• AI prediction runner 구성

🔍 상세

• src/ai/types.ts에 AiPrediction, AiPredictionEvidencePayload, AiPredictionClient, AiPredictionResult 등 AI prediction 관련 타입 추가
• buildAiPredictionEvidencePayload로 BranchRiskAnalysisInput과 deterministic BranchRisk를 AI prediction evidence로 결합
• selectAiPredictionTargets로 deterministic score 기준 prediction 대상 branch 선택
• DEFAULT_AI_PREDICTION_MINIMUM_SCORE로 기본 prediction threshold 상수화
• validateAiPredictionResponse로 AI 응답의 confidence, recommendedActions, falsePositiveNotes schema 검증
• DEFAULT_AI_PREDICTION_SYSTEM_PROMPT와 buildAiPredictionPrompt로 provider 독립 prompt 구성
• predictMergeRisksWithAi로 target selection, prompt 생성, client 호출, response validation 흐름 연결
• branch별 AI prediction 결과를 predicted, skipped, failed 상태로 분리
• 실제 OpenAI provider 연결 없이 AiPredictionClient interface와 fake client 테스트로 core 흐름 검증

✅ 검증

• npm run build
• npm test

📸 영상 / 이미지 (Optional)

없음

[gemini-code-assist]

Code Review

This pull request introduces AI-based merge risk prediction capabilities to the Watcher service, including modules for evidence building, prompt construction, target selection, response validation, and execution coordination, along with comprehensive tests. The reviewer feedback focuses on improving robustness and performance: specifically, parallelizing the AI prediction requests using Promise.all to reduce execution time, and relaxing the schema validation to gracefully handle null or missing fields (such as recommendedActions, falsePositiveNotes, and files) in the LLM responses.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.