Autopilot

OpenClaw is the high-autonomy agent mode. It proactively decides, plans, and executes tasks with deep context fusion, including a full sweep of historical me...

This Autopilot skill is plausible for a high-autonomy agent but contains broad, potentially privacy-invasive instructions (full memory sweep, filesystem scan, listing cron jobs, and auto-executing 'strong' ideas). Before enabling it: (1) confirm what filesystem paths and cron data the agent may access and insist on scoping that to known app directories rather than whole filesystem; (2) require explicit, per-action user confirmation for any automatic execution of proposed ideas; (3) restrict or redact highly sensitive memory files (financial, secrets) from the memory corpus or run the skill in a sandboxed environment; (4) ask the skill author to replace vague steps like "scan filesystem" with precise, limited reads, or add explicit consent prompts; (5) enable audit logging or ask for a preview step so the agent shows planned actions before executing. Those mitigations will reduce privacy/exfiltration risk while keeping the skill's intended autonomy.

Type: OpenClaw Skill Name: openclaw-autopilot-v9 Version: 9.0.1 The skill bundle defines a high-autonomy 'Autopilot' mode (v9) designed for proactive task execution, context-aware planning, and 'Generative Intelligence.' It utilizes a complex framework involving 'Memory Mining' (reading MEMORY.md and memory/*.md), 'Editorial Intelligence' (refining and cutting redundant content), and 'Constraint Alchemy' to handle user requests autonomously. While the agent is granted broad access to historical context and the filesystem, the instructions include explicit safeguards to respect privacy boundaries in shared contexts and adhere to system safety rules. The bundle emphasizes 'Transparent Execution' through a detailed status symbol system and lacks any indicators of malicious intent, data exfiltration, or unauthorized remote execution.