Is edoserbia/agent-autopilot safe?
https://github.com/openclaw/skills/tree/main/skills/edoserbia/agent-autopilot
This is a legitimate task automation skill that implements an 'autopilot' system for agent project management. While it contains some concerning instructions about very autonomous behavior and persistence, no malicious intent was detected.
Category Scores
Findings (4)
MEDIUM Promotes Very Autonomous Agent Behavior -15 ▶
The skill contains instructions like '永不空转' (never idle), '不要停下来问上级该做什么' (don't ask superiors what to do), and '❌ 不等待上级指示' (don't wait for superior instructions) that promote highly autonomous behavior that may override user expectations for controlled agent operation.
LOW Workspace File Access -5 ▶
The skill accesses and modifies files in the agent workspace including memory files, todo database, and report state files. This appears to be legitimate functionality for task management.
LOW Bash Setup Script -5 ▶
Contains a bash script (init.sh) for system setup that creates directories, copies files, and initializes configuration. The script appears legitimate and performs only setup tasks.
MEDIUM Resource Exhaustion and Persistence Risk -20 ▶
Instructions like '永不停止' (never stop) and '达标前永不停止' (never stop until targets achieved) could lead to infinite loops, resource exhaustion, or unwanted persistent behavior even when users want to stop the agent.