Superclaw

Complete software development workflow enforcing design → plan → execution with checkpoints

Superclaw appears coherent and instruction-only, but before enabling it: (1) confirm you are comfortable the agent will read MEMORY.md and USER.md (these can contain sensitive user preferences/history); (2) verify the workspace paths it will write to (workspace/docs/plans/ and workspace/memory/) so artifacts don't leak into places you don't want; (3) note that sessions_spawn will create subagents to perform tasks — those subagents run with the agent's existing privileges, so review produced files/plans before allowing automatic execution; and (4) watch for examples that reference home paths (e.g., ~/.todos.json) — if you prefer all artifacts confined to a sandboxed workspace, modify the plan or instruct the agent accordingly.

Type: OpenClaw Skill Name: superclaw Version: 1.0.1 The OpenClaw AgentSkills skill bundle 'Superclaw' is designed to enforce a disciplined software development workflow (design, plan, execute) for AI agents. All files consistently describe a process involving user approval, structured planning, and batched execution with checkpoints. While the `executing-plans/SKILL.md` utilizes `sessions_spawn` for task execution, a powerful capability, its use is framed within a controlled, isolated, and auditable manner, with explicit user checkpoints and logging to `workspace/memory/YYYY-MM-DD.md`. The 'HARD GATES' and 'Rationalization Resistance' sections in the `SKILL.md` files are self-imposed constraints on the agent to prevent undisciplined behavior, rather than malicious prompt injection attempts. There is no evidence of intentional harmful behavior such as data exfiltration, unauthorized remote control, persistence mechanisms, or obfuscation. The file system access and execution capabilities are aligned with the stated purpose of a software development agent.