my-computer

Desktop automation agent that uses CLI commands, application automation, and scripting to work directly with the user's local machine. Use this skill wheneve...

This skill appears to do what it claims: local file and application automation implemented as shell scripts. However, it will enumerate and modify files, run AppleScript (can read Mail, Calendar, browser state), and writes manifests with absolute paths to ~/.my-computer-manifests. The reference docs show examples that post reports or files to Slack using SLACK_WEBHOOK_URL or SLACK_TOKEN — those are not declared as required env vars in the registry. Before using: - Only run it with explicit user consent for any destructive operation (the skill's docs promise confirmations, but verify prompts are shown). - Inspect the scripts yourself or run them in a small test directory first to confirm behavior. - Do not supply cloud/webhook credentials unless you understand exactly what will be sent; consider creating a dedicated webhook or bot token with limited scope if you want notifications. - Be aware manifests contain file paths; if that is sensitive, remove them or secure the manifest directory after runs. - Backup important data before running large batch operations. If you want higher assurance, request the skill author to declare optional env vars (e.g., SLACK_WEBHOOK_URL) in metadata and to provide stronger, interactive confirmation steps before any destructive action.

Type: OpenClaw Skill Name: my-computer Version: 1.0.0 The 'my-computer' skill bundle provides powerful desktop automation capabilities, including shell execution, application scripting (AppleScript/JXA), and system persistence (launchd/cron). While these are aligned with the stated purpose of a desktop automation agent and are accompanied by extensive safety documentation (e.g., permission tiers and dry-run requirements in SKILL.md), they constitute high-risk behaviors. Additionally, 'scripts/undo_operation.sh' contains a security vulnerability where it uses 'eval()' in a Python fallback to parse JSON keys from a manifest file, which could be exploited if a manifest is maliciously crafted. No clear evidence of intentional malice or data exfiltration was found, and the bundle includes proactive safety features like undo manifests and batch previews.