Graph Memory

Security checks across malware telemetry and agentic risk

Overview

Graph Memory is a coherent memory plugin, but it automatically stores conversations, sends derived content to configured AI providers, and reuses memories across sessions, so users should review its data-handling carefully.

Use this only if you are comfortable with a persistent memory engine for your OpenClaw conversations. Before installing, decide whether conversations may contain secrets, choose trusted LLM/embedding endpoints, verify where the SQLite database is stored, and look for a way to clear or scope stored memories.

VirusTotal

VirusTotal engine telemetry is currently stale for this artifact.

View on VirusTotal

Risk analysis

Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.

#
ASI06: Memory and Context Poisoning
Medium
What this means

Sensitive details or bad instructions from old conversations could persist and influence future agent behavior across sessions.

Why it was flagged

This shows automatic persistent storage of all conversation messages and automatic reuse in later prompts.

Skill content
Message in → ingest ... All messages saved to gm_messages ... turn_index continues from DB max (survives gateway restart) ... Next session → before_prompt_build ... inject into context
Recommendation

Install only if you want persistent cross-session memory; avoid sharing secrets, review the database location, and look for or request clear delete, retention, opt-out, and session-scoping controls.

#
ASI07: Insecure Inter-Agent Communication
Low
What this means

A chosen provider may receive parts of your conversations or derived memory content.

Why it was flagged

The plugin sends conversation-derived content to a configured LLM-compatible provider for extraction and summaries.

Skill content
LLM (`config.llm`) — Required. Used for knowledge extraction and community summaries. Any OpenAI-compatible endpoint works.
Recommendation

Use only trusted providers or local endpoints, check provider retention policies, and avoid putting secrets into conversations that will be ingested.

#
ASI03: Identity and Privilege Abuse
Low
What this means

Misconfigured or overbroad API keys could incur cost or send data to an unintended provider.

Why it was flagged

The README requires provider credentials or an environment-variable fallback, even though registry metadata declares no primary credential or required env vars.

Skill content
"apiKey": "your-llm-api-key" ... "apiKey": "your-embedding-api-key" ... falls back to the `ANTHROPIC_API_KEY` environment variable
Recommendation

Use separate least-privilege keys where possible, store them securely, and verify the configured baseURL before enabling the plugin.

#
ASI04: Agentic Supply Chain Vulnerabilities
Low
What this means

Running the installer gives an external binary local ability to modify OpenClaw setup.

Why it was flagged

The optional Windows installer is a downloaded executable that changes local OpenClaw configuration and restarts a service.

Skill content
Download `graph-memory-installer-win-x64.exe` ... it auto-detects your OpenClaw installation ... configures `plugins.slots.contextEngine`, adds the plugin entry, and restarts the gateway
Recommendation

Prefer the package-manager install path when possible, or verify the release source/signature before running the Windows installer.