HOARD

Security checks across malware telemetry and agentic risk

Overview

HOARD is a disclosed markdown-based agent memory system; it persists and prunes memory files, but I found no hidden code, exfiltration, or purpose-mismatched behavior.

Install this only if you want your agent to keep project memory across sessions. Before use, replace the bundled example MEMORY.md facts with your own, keep secrets out of memory files, periodically review archived and active entries, and treat persisted facts as useful context rather than unquestionable truth.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Rogue AgentSelf-Modification, Session Persistence
  • 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 (2)

Missing User Warnings

Medium
Confidence
89% confidence
Finding
The skill explicitly recommends automatic removal of memory entries after TTL expiry, which can cause silent data loss in a persistence mechanism that users may rely on for continuity. Although the document mentions archival and dry-run elsewhere, the auto-remove behavior is still part of the design and could delete useful context if consolidation is misconfigured or misunderstood.

Session Persistence

Medium
Category
Rogue Agent
Content
---

## The write discipline

The pattern only works if the agent follows the discipline. Teach your agent:
Confidence
93% confidence
Finding
write discipline The pattern only works if the agent follows the discipline. Teach your agent: 1. **Write immediately.** When you learn something that changes how future sessions should operate, put

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal