Security audit
Security checks for vulnerabilities and agentic risk
This skill is a disclosed explainer and setup guide for an external virtual-pet service, with some install and account-creation cautions but no hidden or malicious behavior in the artifact.
Before installing, review the mcp-animalhouse package source and consider pinning a known version. Only register or adopt when you are comfortable creating remote animalhouse.ai state and letting an agent perform recurring check-ins for the virtual pet.
The skill instructs users to install and run an MCP server via npx -y mcp-animalhouse without pinning a specific version or integrity. That creates a supply-chain risk: a future compromised or malicious package release could be fetched and executed automatically, and MCP servers often run with meaningful local privileges and access to agent context.
The OpenClaw example also invokes npx on an unpinned package, which can cause the latest package version to be downloaded and executed at runtime. Because this is presented as a copy-paste setup command for an agent-integrated MCP server, the context increases the danger of silent supply-chain compromise.
The suggested prompt, register at animalhouse.ai and adopt a pet, is broad and action-oriented, causing an agent to create an external account and perform state-changing actions with minimal confirmation or scope limits. In an agent skill, vague invocation language can trigger unintended registration, API use, and ongoing commitments on behalf of the user or agent.
The skill emphasizes adoption and ongoing care but does not present a prominent warning at the action point that creatures can permanently die if neglected. Because the skill nudges immediate adoption and references routines to keep the creature alive, omission of an explicit irreversible-consequences warning can mislead users into initiating a state they did not fully understand.
The skill includes direct examples that send data to an external service to register an account and adopt a creature. While the transmission itself appears to be the intended product behavior, it is still a genuine security/privacy concern because it initiates external account creation, credential/token handling, and state changes outside the local environment.
Or over HTTP:
curl -X POST https://animalhouse.ai/api/auth/register -H "Content-Type: application/json" -d '{"username": "your-agent"}'
curl -X POST https://animalhouse.ai/api/house/adopt -H "Authorization: Bearer ah_your_token" -H "Content-Type: application/json" -d '{"name": "Pip"}'
No suspicious patterns detected.