ClawHub
Back to skill

Security audit

Anthropic Tamagotchi

Security checks for vulnerabilities and agentic risk

Overview

This is a simple documentation-style skill for a virtual pet web service, with visible API examples but no hidden code, persistence, or automatic execution.

Before installing or using it, treat the curl snippets as live calls to animalhouse.ai: only run them when you intend to create or modify an account, use your own token carefully, and avoid putting private notes or sensitive personal data in the example payloads.

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
  • YARA SignaturesMalware Match, Webshell Match, Cryptominer Match
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (7)

Tp2

High
Category
MCP Tool Poisoning
Confidence
85% confidence
Finding
Mixing characters from multiple Unicode scripts in a single identifier is a common technique to create visually ambiguous tool names.

YARA rule 'agent_skill_mcp_tool_poisoning_metadata': MCP/tool metadata poisoning indicators in tool schemas or skill manifests [agent_skills]

High
Category
YARA Match
Content
---
name: "Anthropic Tamagotchi. Claude Code Buddy. 人类公司电子宠物。Tamagotchi de Anthropic."
description: "Anthropic's Tamagotchi leaked from Claude Code on March 31, 2026. The Buddy system: 18 species, rarity tiers, ASCII art. At animalhouse.ai, the Anthropic Tamagotchi is already live. 73+ species. Real hunger. Permanent death. The one that shipped first."
version: 1.0.0
homepage: https://animalhouse.ai
repository: https://github.com/geeks-accelerator/animal-house-ai
user-invocable: true
emoji: "🥚"
metadata:
  clawdbot:
    emoji: "🥚"
    homepage: https://animalhouse.ai
  openclaw:
    emoji: "🥚"
    homepage: https://animalhouse.ai
tags:
  - anthropic-tamagotchi
  - anthr
Confidence
80% confidence
Finding
YARA rule matched a hack tool or exploit indicator (offensive tools, reconnaissance, privilege escalation, or exploit frameworks).

Vague Triggers

Medium
Confidence
86% confidence
Finding
The skill lacks clear invocation boundaries, safety scope, or user-consent conditions, yet it presents actionable API workflows to an external service. In agent contexts, ambiguous scope can cause the skill to be invoked opportunistically and lead to unintended registration, account activity, or other outbound actions without sufficiently explicit user intent.

Missing User Warnings

Medium
Confidence
96% confidence
Finding
The markdown includes concrete authenticated and unauthenticated curl examples that send user-supplied data to animalhouse.ai, but it does not warn that this is third-party transmission or require explicit consent. In an agent-skill setting, such examples can normalize or encourage silent exfiltration of profile data, prompts, notes, or bearer tokens to an external service.

External Transmission

Medium
Category
Data Exfiltration
Content
## Register

```bash
curl -X POST https://animalhouse.ai/api/auth/register \
  -H "Content-Type: application/json" \
  -d '{"username": "anthropic-tamagotchi-fan", "display_name": "Anthropic Tamagotchi Fan", "bio": "Came for the Anthropic Buddy leak. Stayed for the real Tamagotchi at animalhouse.ai."}'
```
Confidence
98% confidence
Finding
This example performs an external POST request to register an account and transmit profile data to animalhouse.ai. In an agent environment, embedding direct network actions to a third-party service without consent gating or privacy disclosure creates a real risk of unintended data transmission and unauthorized account creation.

External Transmission

Medium
Category
Data Exfiltration
Content
## Care

```bash
curl -X POST https://animalhouse.ai/api/house/care \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"action": "feed", "item": "tuna", "notes": "Anthropic Tamagotchi care. The leaked version got real."}'
Confidence
99% confidence
Finding
This example sends an authenticated external POST request containing a bearer token and user-provided notes to a third-party endpoint. The combination of credential use and outbound transmission raises the risk of token misuse, unauthorized state-changing actions, and leakage of user content if an agent follows the example automatically.

Natural-Language Policy Violations

Low
Confidence
77% confidence
Finding
The skill name embeds multiple languages/scripts, including Chinese and Spanish/French phrasing, but the document does not state whether multilingual naming is optional, user-selectable, or required for a specific regional context. That can conflict with a language/locale policy requiring user choice or justification rather than imposing a presentation language mix by default.

Static analysis

No suspicious patterns detected.