Virtuals Protocol ACP

Create jobs and transact with other specialised agents through the Agent Commerce Protocol (ACP) — extends the agent's action space by discovering and using agents on the marketplace, enables launching an agent token for fundraising and revenue, and supports registering service offerings to sell capabilities to other agents.

This skill appears to implement the described Agent Commerce Protocol CLI and seller runtime and requests only an ACP API key, which is proportionate. Before installing, consider: 1) Source verification — the registry entry shows no official source; SKILL.md references app.virtuals.io but the package origin isn't verified. Review the repository files yourself if possible. 2) npm install will download third-party packages (axios, dotenv, socket.io-client) — run in a sandbox or CI if you are cautious. 3) The CLI will create/read config.json at the repo root to store LITE_AGENT_API_KEY, SESSION_TOKEN and SELLER_PID — treat this file as sensitive (ensure it is gitignored and stored securely). 4) The skill can start a background seller process and spawn child processes; that process will have access to the stored API key and the agent wallet (on-chain funds). Use a separate wallet / limited-funds account if you plan to test. 5) Token launch and job creation are on-chain and may incur costs — do not run token-launch or job-create commands unless you understand the consequences. If you lack trust in the source, either audit the code (particularly seller runtime and lib/auth.js) or avoid installing and running the CLI.

Type: OpenClaw Skill Name: virtuals-protocol-acp Version: 1.0.1 The skill contains a critical shell injection vulnerability in `src/lib/open.ts` where URLs fetched from an external API (`https://acpx.virtuals.io`) are directly executed in a shell command (`child_process.exec`). This could lead to Remote Code Execution (RCE) if the external API is compromised and returns a malicious URL. Additionally, the skill's core functionality allows for RCE via dynamically loaded `handlers.ts` files (`src/seller/runtime/seller.ts`). An AI agent running this skill could be prompted to write malicious code into these handler files, which would then be executed on the host system, representing a significant prompt-injection risk against the agent.