Security audit

Hermes Learning Loop2

Security checks for vulnerabilities and agentic risk

Overview

The skill appears to be a memory/self-improvement helper, but it asks the agent to persist broad task details and user corrections across sessions without clear consent, redaction, or retention limits.

Review this before installing if you handle confidential work, personal data, credentials, or customer information. Only use it where persistent memory is intended, and configure or enforce rules to store sanitized summaries, avoid secrets and private details, require user consent for sensitive notes, and provide a way to delete or revise retained memories.

SkillSpector

Vulnerability Patterns

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

Findings (3)

Vague Triggers

Medium

Confidence: 80% confidence
Finding: The activation rule 'completed a complex task' plus similar criteria is ambiguous and can trigger this skill after many ordinary interactions without explicit user consent. In context, activation leads to writing logs and updating memory files, so over-broad invocation increases the chance of retaining unnecessary or sensitive task details across sessions.

Ssd 3

Medium

Confidence: 96% confidence
Finding: The skill directs the agent to keep daily logs of 'all valuable operations' and lessons learned in persistent memory files. That creates a clear data-retention risk: sensitive prompts, secrets, proprietary workflows, or personal data may be stored in natural language and later resurfaced, searched, or reused outside the original context.

Ssd 3

Medium

Confidence: 97% confidence
Finding: This section mandates recording user corrections and propagating them into memory and related knowledge stores. User corrections often contain sensitive context, exact commands, internal procedures, or private preferences; persisting them across sessions can leak confidential information and cause unintended resurfacing in future tasks, making the context more dangerous than a generic note-taking feature.

Static analysis

No suspicious patterns detected.