Solo Impl

Autonomous executor for Solo CLI — actually runs setup, calibration, teleoperation, dataset recording, policy training, and inference commands in the user's...

This skill appears to do what it says (run Solo CLI flows) and only asks for conditional tokens for HuggingFace/W&B when you opt into those features. Before installing or using it: 1) Verify the skill's source (registry shows 'Source: unknown' and skill.json/version mismatch) and confirm you trust the referenced GitHub repositories. 2) Be cautious about the uv installer step: it pipes a remote script (https://astral.sh/uv/install.sh) to sh — only proceed if you trust astral.sh; consider installing uv yourself and answering 'no' to the consent prompt. 3) For pushes to HuggingFace or W&B, prefer scoped or temporary tokens and understand the skill will use whichever credentials are present. 4) When the skill prints the full command before opening a terminal, review it carefully (it will open terminals and run commands immediately after pre-flight). 5) If you have low tolerance for an agent executing system-level commands on your machine, do not enable autonomous invocation for this skill or run the steps manually instead.

Type: OpenClaw Skill Name: solo-impl Version: 1.3.0 The skill bundle provides an autonomous executor for the Solo robotics CLI, requiring high-privilege shell access to perform system setup and hardware control. It employs several high-risk execution patterns, most notably a 'curl | sh' installer for the uv package manager (domains/environment.json) and the use of osascript or gnome-terminal to spawn new terminal windows with dynamically constructed command strings (prompts/impl_executor_prompt.txt). While these capabilities are plausibly necessary for the stated purpose and the bundle includes 'Transparency Rules' to disclose actions to the user, the autonomous nature of the agent and the potential for shell injection via unsanitized parameters (e.g., task descriptions in record_dataset) represent a significant attack surface. No evidence of intentional malice, such as data exfiltration or backdoors, was detected.