OpenClaw agents feel useful because they evolve files about themselves (AGENTS, SOUL, IDENTITY, etc).
OpenGoat's goal is that agents also evolve files about an organization (MISSION, VISION, STRATEGY, KPIs, ROADMAP).
What it does: * Define an org hierarchy where agents have direct reports * Executives and managers plan and delegate tasks to their teams * Individual agents execute tasks and update status and artifacts * Tasks can be generated from a top-down strategy (default) or an optional bottom-up mode
Another key feature: Specialist agents can run on other runtimes, not just OpenClaw (for example Claude Code, Codex, Cursor, OpenCode).
I started this as an experiment to see what happened if we let AIs operate more freely and take more decisions themselves.
A few interesting facts I've seen while running the tool:
1. Agents created markdowns for culture and values, even though they were never prompted to do it. 2. Agents respond differently if a task comes from a peer than from a manager. If it comes from a manager, they follow it like they follow direct prompts. 3. In one occasion, an agent sent a complaint about his manager to higher management.
I'd love to hear some critical feedback.