Skip to content

Security: harshitboots/multi-agent-data-pipeline

Security

SECURITY.md

Security Policy

Supported Versions

Only the latest release on the main branch is supported with security fixes.

Version Supported
1.2.x
< 1.2

Reporting a Vulnerability

Please do not open a public GitHub issue for security vulnerabilities.

Instead, report privately via:

Include a description of the issue, steps to reproduce, and the potential impact. You should expect an initial response within 5 business days.

Scope

This project handles two categories of sensitive data:

  • API credentials — Anthropic API keys (via .env, st.secrets, or BYOK in the sidebar) and database connector credentials (Databricks, Snowflake, Postgres, MySQL, BigQuery, Redshift).
  • User-uploaded data — CSV/PDF files may contain PII, which the PII Anonymiser agent is designed to detect and mask before downstream processing.

Reports in these areas are especially welcome:

  • Ways a BYOK key or connector credential could leak into logs, SQLite (pipeline_runs.db), or the observability dashboard.
  • Ways PII could bypass the anonymiser and reach an LLM prompt or the run history.
  • Injection risks in database connector queries (e.g. src/connectors/*.py table-name interpolation).
  • Bypasses of the VPN/proxy access-control block.

Known Design Notes (not vulnerabilities)

  • BYOK API keys are stored only in st.session_state for the browser session — never persisted to disk or SQLite.
  • Anonymous run tracking stores a SHA-256 hash of IP + User-Agent, never the raw IP.

There aren't any published security advisories