Security audit
memory-vector-bge
Security checks for vulnerabilities and agentic risk
Overview
The skill's files are largely consistent with a local memory-distillation and vector-search tool, but there are mismatches and privacy/exfiltration risks (hard-coded personal data, file overwrite behavior, and network calls to configurable endpoints) that warrant caution before installing.
Before installing or running this skill: - Backup your workspace, especially any existing MEMORY.md and memory/ files — the distill script will write/overwrite memories.json and MEMORY.md. - Review and, if needed, remove the hard-coded personal data in updateMemoryCore inside dist/memory-distill.js (it currently writes specific family/address lines). - Only set EMBEDDING_URL and LLM_URL (and any API keys) to endpoints you control or fully trust. The scripts will POST full log content to those URLs; untrusted endpoints can exfiltrate sensitive data. - Be aware the provided 'sensitive filter' is simple regex-based and can miss secrets; do not rely on it for protecting high-risk credentials. - Note the code uses CONFIG.embedding.apiKey in the HTTP Authorization header for all requests (including LLM calls) — if you plan to use an authenticated LLM endpoint, check/patch the code so the correct API key is sent. - If you lack confidence in the code, run it in an isolated environment (container/VM) and inspect logs/network traffic first. If you want, I can: (a) point out the exact lines to change to avoid overwriting MEMORY.md or remove hard-coded personal info, (b) propose a safe patch to use per-service Authorization headers, or (c) produce a minimal audit checklist for deployment.
Static analysis
No suspicious patterns detected.