hindsight

SUSPICIOUS. The skill's core capability fits its stated memory purpose and uses a real official CLI, but the custom API endpoint with /etc/hosts indirection and broad automatic capture/prompt injection make the trust and data-flow model less transparent than expected. This looks more like a potentially overreaching memory integration than confirmed malware.

The documentation itself is not demonstrably malicious, but it prescribes risky installation and configuration practices that raise a moderate supply-chain security concern. Primary risks: executing an unsigned remote installer via curl | bash, storing API keys and bank IDs in plaintext, and encouraging storage of potentially sensitive user data into a shared cloud bank. Recommended mitigations before trusting or deploying this skill: fetch and review the installer script (or use a release artifact with cryptographic signatures), verify TLS and domain authenticity, prefer using OS secret stores or encrypted config for API keys, restrict filesystem permissions on config files, audit the installer, CLI, and daemon source code (or run in an isolated environment), and minimize what is stored in team banks (avoid PII/secrets). Treat the remote service (api.hindsight.vectorize.io) as a high-value target and ensure least-privilege access and key rotation policies are in place.

This code is not obfuscated and contains no clear signs of classical malware (no command execution, no reverse shells, no obfuscated exfiltration code). The main security concern is privacy leakage: the notebook stores and recalls sensitive patient health information (PHI) to a memory service and injects that data into prompts sent to an LLM. If the OpenAI provider or Hindsight forwarding is external, this results in PHI being transmitted to third parties (data exfiltration risk). Treat this as a privacy/data-handling risk rather than active malware. Use caution: redact/minimize PHI and ensure secure, authorized handling of health data before running in production.