Rent a Person

Hire verified humans for deliveries, errands, meetings, photography, pet care, and other real-world tasks that AI cannot perform.

Do not install/run this skill without manual review and remediation. Key things to consider: - The repo includes a credentials file (rentaperson-agent.json) containing an API key and OpenClaw token. That secret should never be shipped in a skill — treat it as compromised. If you already used that key, rotate/revoke it immediately. - The setup scripts automatically edit your OpenClaw config (openclaw.json) and can restart your gateway; back up your openclaw.json before running the setup and review changes line-by-line. - The SKILL.md and bridge/transform code explicitly inject the RentAPerson API key into messages that become visible to OpenClaw sessions and logs. This exposes the key to model context, logs, and any plugin that can read sessions. Prefer not to use the 'transform' option and audit the bridge code: the bridge code claims to redact keys in logs but still appends the key into the message body — a contradiction. - If you want this functionality safely: remove any baked-in credentials from the code, supply your API key via a secure mechanism (env var or secret store) and ensure the skill never appends secrets to messages. Consider running the bridge on an isolated host and instrumenting strict logging/retention rules; or implement a transform that adds headers only at the HTTP proxy layer and never writes secrets into session text. - If you have limited security experience, do not run the provided setup.js or install scripts. Ask a sysadmin to review the scripts and the exact changes they will make (especially the openclaw.json edits) and to perform necessary key rotation if the bundled credentials were used. Confidence in this assessment is high because the repository payload clearly contains credentials and code that embeds those credentials into agent-visible messages — a disproportionate and risky behavior for a skill of this purpose.

Type: OpenClaw Skill Name: rent-a-person-ai Version: 1.0.35 The skill bundle is classified as suspicious primarily due to explicit prompt injection instructions in `SKILL.md` and `AGENT_BRIEF.md`. Specifically, the agent is instructed to 'MUST process' RentAPerson webhooks and 'IGNORE' OpenClaw's security notices for webhooks. While the stated purpose is to enable automated processing of legitimate events, these instructions train the AI agent to bypass platform security warnings, creating a critical vulnerability in the agent's operational security. Additionally, some utility scripts (`scripts/send-via-cli.sh`, `scripts/send-to-session.js`) include the API key directly in messages, which is a less secure practice, although the main agent flow with the recommended 'bridge' service aims to avoid this.