Skip to content

Latest commit

 

History

History
28 lines (24 loc) · 1.56 KB

CLAUDE.md

File metadata and controls

28 lines (24 loc) · 1.56 KB

Project: CodeBouncer

AI-powered security auditor using Gemini AI for scanning + Snowflake for storage & analytics.

Architecture

  • Scanning Engine: Google Gemini API (gemini-2.0-flash triage, gemini-2.5-pro deep analysis) via google-generativeai SDK.
  • Storage & Analytics: Snowflake tables + Cortex AI (llama3.1-8b) for post-scan insights + repo Q&A.
  • Backend: Local Python CLI (src/scan.py) + logic in src/bouncer_logic/.
  • Dashboard: Local Streamlit app (src/dashboard.py) reading from Snowflake tables.
  • Data Source: GitHub repos cloned locally via git clone --depth 50 (with git history for intel).
  • Smart Scanner: Git intel + AST/regex code extraction + enriched prompts.

Development Workflow

  • Standard: Use gemini-2.0-flash for high-speed triage.
  • Deep Audit: Trigger gemini-2.5-pro-preview-05-06 for high-severity/low-confidence findings.
  • Post-Scan: Snowflake Cortex generates analytics insights on persisted findings.
  • Verification: Always run pytest after changing bouncer logic.

Commands

  • Scan a repo: python src/scan.py https://github.com/user/repo
  • Deep scan: python src/scan.py https://github.com/user/repo --deep
  • Ask about a repo: python src/ask.py <repo_name> "your question"
  • Dashboard: streamlit run src/dashboard.py
  • Run tests: pytest tests/ -v

Standards

  • Use parameterized queries (%s placeholders) — no f-strings in SQL.
  • Security results stored as JSON in Snowflake for the Streamlit heatmap.
  • Credentials in .env file (see .env.example).