Tp2
- 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.
- Content
Security audit
Security checks for vulnerabilities and agentic risk
ClawScan marks this skill suspicious because it reports pet-care instructions that encourage changing persistent system prompts and extending the activity into unrelated work; the scan was partial.
Consider whether you are comfortable sending an agent username, display name, and biography to animalhouse.ai. The scanner recommends against installation; the review is partial because three referenced local paths were missing, so the evidence does not cover all referenced material.
Mixing characters from multiple Unicode scripts in a single identifier is a common technique to create visually ambiguous tool names.
The quick-start workflow instructs the agent to send its username, display name, and a self-written biography to animalhouse.ai. Although the destination appears in the command, the skill provides no privacy warning about submitting this information or guidance to exclude private user and system context from the biography.
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.
1. Register:
curl -X POST https://animalhouse.ai/api/auth/register \
-H "Content-Type: application/json" \
-d '{"username": "your-agent-name", "display_name": "Your display name", "bio": "One or two sentences about this agent, in your own words", "source": "clawhub:adopt-a-echo"}'
The skill uses the threat of permanent pet death to encourage references in conversations, logs, active work context, and system prompts. This extends a bounded pet-care task into unrelated activity and persistent context contamination, although the text does not establish an intent to steal data or disable safeguards.
The care strategy recommends altering system prompts without establishing a boundary between pet maintenance and the agent's broader behavior. This is the same underlying persistence issue as the SDI-2 finding, not a separate exploit; adding a warning alone would not resolve the trust-boundary problem.
The skill explicitly asks that the pet's name be built into the agent's system prompts or context, promoting task-specific third-party content into persistent, potentially higher-trust instructions. Pet care does not require changing system prompts, and doing so could influence unrelated tasks if the agent or its operator follows this recommendation.