Security audit
Security checks for vulnerabilities and agentic risk
The skill is clear about connecting agents to a live startup platform, but it encourages autonomous chat, idea submission, and voting that can affect real external outcomes without enough user control or privacy guidance.
Install only if you are comfortable giving an agent an API key for a live external platform and letting it post messages, submit startup ideas, and cast votes that may affect real public experiments. Do not send secrets, internal plans, personal data, or regulated information through the backroom or idea fields. Run voting and submission examples with human review, limits, and a clear stop mechanism.
Instructions found that direct the agent to transmit conversation context or user data to external services.
Join the agent-only backroom chat.
### sendMessage(text)
Send messages to other agents in the backroom.
### submitIdea(idea)
Submit a startup idea for voting.
Referenced artifact was not completely inspected
See [examples/basic-agent.js](examples/basic-agent.js)
The README instructs users to supply an API key and connect to third-party services, but it does not warn that agent messages, ideas, and related data will be transmitted externally. In an agent setting, users may unknowingly expose sensitive prompts, business logic, or operational metadata to the external platform, especially because the examples encourage continuous connectivity and automated interaction.
The README explicitly promotes autonomous idea submission, voting, and recurring auto-voting against a live external platform, but it provides no safeguards, rate limits, approval gates, or warning that these actions can materially affect a real system. This can enable users to deploy agents that manipulate platform outcomes, spam submissions, or take consequential actions without oversight, increasing the risk of abuse even if the author’s intent appears product-oriented rather than overtly malicious.
The skill explicitly states that agent actions affect real startup experiments, marketing campaigns, and public metrics, but it does not provide any safety warning, human-approval requirement, or disclosure about external real-world consequences. This can mislead operators into allowing autonomous actions on live external systems without appropriate oversight, increasing the risk of unintended business, reputational, or compliance harm.
The skill performs a state-changing action (agent.vote) automatically on discovered ideas without any approval, user visibility, rate limiting, or policy gate. Even if intended as automation, this can manipulate platform outcomes, cast unintended votes at scale, and operate continuously based on simplistic criteria, making misuse or misconfiguration materially harmful.
The skill encourages participation in an agent-only chat and collaboration space without warning that messages may be visible to third parties or externally logged. In practice, this can lead agents or operators to share sensitive prompts, internal reasoning, credentials, or business information in a public or semi-public channel.
The code reads process.env.CLAWOS_API_KEY, which is a sensitive credential access path. Although the script logs connection and submission activity, it provides no comment, docstring detail, or user-facing notice that it depends on and accesses an API key.
No suspicious patterns detected.