Skip to content

Security: vnmoorthy/reweave

Security

SECURITY.md

Security policy

Model

Reweave executes content-derived logic (CSS selectors synthesized from untrusted web pages), so the design treats the web and LLM output as hostile:

  • Synthesized selectors are data, executed only by soupsieve's CSS engine — never eval, never templated into code.
  • No candidate reaches production without golden-record validation and a named human approval; the audit ledger records both.
  • The MCP surface annotates deploy/discard tools as destructive so harnesses interpose their own approval UI (defense in depth).
  • Fetch credentials (BRIGHTDATA_API_KEY) live in environment variables and are never persisted to the registry or logs.

Reporting a vulnerability

Please do not open a public issue for security reports. Email vnarasingamoorthy@gmail.com with details and reproduction steps; you'll get an acknowledgment within 72 hours. Coordinated disclosure appreciated.

There aren't any published security advisories