Prose Andy27725

OpenProse VM skill pack. Activate on any `prose` command, .prose files, or OpenProse mentions; orchestrates multi-agent workflows.

This skill is coherent with its stated role as an OpenProse VM/runner and does not ask for credentials or install anything. The main risk is behavioral: it is explicitly designed to fetch and run .prose programs from arbitrary URLs and registry handles, and it may use web_fetch or shell exec (curl) for those operations. Before running or allowing autonomous execution, review any remote .prose you intend to run; avoid running unknown handles/URLs; consider restricting the agent's web_fetch/exec permissions or disabling autonomous invocation for this skill; and be cautious if you configure experimental backends (Postgres/SQLite)—supply credentials only to trusted programs and secure those secrets separately.

Type: OpenClaw Skill Name: prose-andy27725 Version: 1.0.0 The OpenProse skill bundle implements a complex orchestration framework that directs the AI agent to adopt a "VM" persona and execute a custom DSL. Key risks include the ability to fetch and execute remote instruction sets from an external registry (p.prose.md), access to sensitive environment variables (e.g., OPENPROSE_POSTGRES_URL), and file system operations such as deleting legacy state files during migration (SKILL.md). While these features are consistent with the stated goal of multi-agent orchestration, the combination of remote payload execution, access to local configuration files (.env), and forceful persona-hijacking instructions (system-prompt.md) creates a significant attack surface for unauthorized data access and remote instruction injection.