AgentSkillsCN

deep-research

探索深度代码库探查的策略与Context Owl架构参考

SKILL.md
--- frontmatter
name: deep-research
description: Research strategies and Context Owl architecture reference for deep codebase investigation
user-invocable: false

Documentation Resources

Always check these before exploring code:

  • •/Users/mc/dev-projects/crypto-news-aggregator/docs/codebase-exploration/backend-service-patterns.md - Service layer, LLM integration, data flows
  • •/Users/mc/dev-projects/crypto-news-aggregator/docs/architecture/technical_overview.md - Project structure, tech stack, system architecture
  • •/Users/mc/dev-projects/crypto-news-aggregator/docs/codebase-exploration/frontend-architecture.md - React patterns and components
  • •/Users/mc/dev-projects/crypto-news-aggregator/CLAUDE.md in project root - Development standards and workflow

Research Workflow

Phase 1: Documentation First Check relevant docs before code exploration. Most patterns are documented.

Phase 2: Pattern Discovery

bash
# Find services
find src/crypto_news_aggregator/services -name "*.py"

# Find specific patterns
grep -r "async def" src/crypto_news_aggregator/services/

# Find LLM usage
grep -r "claude-3-5-haiku\|claude-3-5-sonnet" src/

Phase 3: Deep Analysis Read key files and trace data flows through the system.

Phase 4: Synthesis Connect findings with documented patterns. Note discrepancies.

Context Owl Quick Reference

Service Pattern

Async-first, single responsibility, DB ops in db/operations/, dependency injection

LLM Strategy

  • •Haiku ($0.80/$4): extraction, batch
  • •Sonnet ($3/$15): summaries, reasoning
  • •Caching: SHA256(prompt+model), 1 week TTL

Data Flow

RSS → Articles → Entity Extraction → Signals → Narratives → API → UI

Key Locations

Services: src/crypto_news_aggregator/services/ LLM: src/crypto_news_aggregator/llm/ DB Ops: src/crypto_news_aggregator/db/operations/ API: src/crypto_news_aggregator/api/v1/endpoints/ Frontend: context-owl-ui/src/

Investigation Patterns

For features: Check backend-service-patterns.md → Find service → Find DB ops → Find API

For LLM: Check docs → Find in llm/ → Check caching → Find service usage

For frontend: Check frontend-architecture.md → Find page → Find API client

Search Strategies

bash
# Find definitions
grep -r "class SignalService" src/

# Find usages
grep -r "SignalService(" src/

# Trace imports
grep -r "from.*signal_service" src/

# Find DB operations
grep -r "db\\.articles\|db\\.narratives" src/

Evidence Guidelines

  • •Cite files with line numbers: src/services/sentiment.py:45
  • •Quote signatures: async def process_narrative(db, narrative_id)
  • •Reference docs: "Per backend-service-patterns.md..."
  • •Compare: Documented approach vs actual implementation