Zelda Agent Memory System
Security checks across static analysis, malware telemetry, and agentic risk
Overview
The skill is broadly coherent as a local agent long-term memory system, but its runtime instructions encourage scanning and recording references to external credentials/paths and workspace files (including paths outside the workspace) while declaring no safeguards — this mismatch and some undocumented behaviors are cause for caution.
This skill implements a local memory system and includes scripts that will create and modify files under your OpenClaw workspace (MEMORY.md, memory/, heartbeat-state.json, and generated topic files). Key concerns: (1) the docs encourage recording references to external systems and API key locations (even paths outside the workspace), which could lead to sensitive information being copied into the memory store; (2) the docs describe a workspace-scanning behavior not present in the provided scripts (mismatch between doc and code); (3) the scripts themselves do not exfiltrate data, but they will persist entries derived from logs and workspace content. Before installing, consider: run the scripts in a safe, restricted test workspace (not your real home or credential-containing directories); inspect and/or modify the code to avoid ingesting or writing secrets; restrict the base-path to a directory that cannot contain system credentials; and decide a policy for what memory should never store (tokens, private keys, passwords). If you need the feature but are unsure, ask for a version that explicitly filters/omits credential paths and confirms no reading outside the declared base path.
Static analysis
No static analysis findings were reported for this release.
VirusTotal
VirusTotal findings are pending for this skill version.
Risk analysis
No visible risk-analysis findings were reported for this release.