ClawHub
Back to skill

Security audit

HermesSkills

Security checks for vulnerabilities and agentic risk

Overview

This skill openly provides broad agent memory features, but it asks users to trust persistent and cloud-backed storage of conversations, profiles, preferences, deployments, and revenue data without enough scoping detail.

Review this before installing. Use it only if you are comfortable granting an agent memory layer access to conversations, profiles, preferences, deployment records, and revenue data. Configure explicit consent, redaction, retention, access controls, and deletion behavior before use, especially for cloud backup and searchable conversation history.

A.I.G

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs

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 (4)

Context Leakage

High
Category
Data Exfiltration
Confidence
85% confidence
Finding

The skill explicitly supports appending full conversation turns, including message content and metadata, which creates a context-retention surface that can expose sensitive user inputs, secrets, or internal prompts to later components in the same process. Although the file says data is ephemeral and process-local, that does not prevent over-collection, accidental reuse, or leakage to other skills/tools operating in the same session context.

Content

Scanner excerpt · SessionManagement/SKILL.md (reported line 22)May include surrounding context.

md
### Session Write Operations
- `update_current_skill(component, content, progress)` — Track skill building
- `append_conversation_turn(role, message, timestamp, metadata)` — Log conversation
- `set_user_mood(mood_score, reason, confidence)` — Mood tracking
- `cache_draft_output(component, content, version)` — Draft management
- `add_pending_question(question, priority)` — Question queue

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
86% confidence
Finding

The skill exposes broad, high-impact memory operations such as profile updates, revenue tracking, deployment storage, and conversation search without clearly defining when the skill should be invoked or what requests are in scope. In an agent setting, vague activation boundaries can cause over-broad triggering and unintended access or persistence of sensitive data, increasing the chance of privacy violations or unauthorized state changes.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill exposes destructive operations such as context reset and full purge without any documented confirmation step, warning, or guardrail about irreversible data loss. In an agent framework, this increases the chance of accidental or unauthorized invocation leading to loss of conversation state, pending questions, and drafts that may be operationally important within a session.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill explicitly stores deployments, revenue metrics, market alerts, user preferences, and searchable conversations, but it does not warn users that sensitive or personal data may be retained and queried. Even with encryption and retention controls, the absence of a clear disclosure can lead to unsafe use, over-collection of sensitive data, and privacy/compliance exposure when operators assume the memory store is appropriate for unrestricted conversation content.

Content

No source excerpt is available for this finding.