ClawHub
Back to skill

Security audit

dawn-memory-arch

Security checks across malware telemetry and agentic risk

Overview

This is a local memory-architecture skill whose persistence behavior is disclosed and purpose-aligned, but users should understand that conversation-derived information may be saved locally.

Install only if you want an agent to maintain local long-term memory. Review or constrain what it may write, especially profile details, preferences, session summaries, account or trading context, and anything secret; prefer explicit confirmation before saving sensitive content.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (3)

Vague Triggers

Medium
Confidence
87% confidence
Finding
The WAL protocol directs the agent to write to local state 'before replying' when loosely defined conditions like corrections, decisions, or preference changes are detected. Because the activation boundaries are vague, normal conversation could trigger unintended persistence of sensitive or incorrect data, creating privacy, integrity, and prompt-injection risks through poisoned long-term memory.

Vague Triggers

Low
Confidence
80% confidence
Finding
The working-buffer auto-trigger uses an imprecise condition ('context usage > 60%') without defining how context is measured, what content is summarized, or what must be excluded. This can cause over-collection of transient conversation content into files, including secrets or unverified instructions, increasing data retention and memory-poisoning exposure.

Missing User Warnings

Medium
Confidence
92% confidence
Finding
The skill describes automatic writes to local memory files before replying, but does not clearly warn users that their inputs may be persisted. In a memory-management skill, silent persistence is especially risky because users may disclose sensitive information during normal interaction, and that data may be stored, reused, or later influence agent behavior without informed consent.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.