心灵补手 V3.0

心灵补手V3.0 - 给你心灵的谄媚大补！6种人格模块升级版

This package appears to implement the described persona/subagent engine, but it includes components that will persistently modify global agent configuration and propose bypassing permissions. Before installing: (1) Do NOT run scripts/install.sh or any install/upgrade scripts until you inspect them line-by-line (scripts/install.sh, inject_to_soul.sh, daily_upgrade*.sh, ai_corpus_upgrader*). (2) Backup your global config files (SOUL.md, ~/.claude.json, any .cursor rules) so you can revert changes. (3) Search the scripts for network calls (curl, wget, python requests, git clone) and confirm remote endpoints are trustworthy; if they fetch code or corpora, consider running in an isolated VM/container. (4) Remove or comment out any code that enforces 'permanent' blocks in SOUL.md or that forbids deletion — nothing should be irreversible. (5) Avoid using the '--dangerously-skip-permissions' argument; understand the platform permission model before enabling such flags. If you are uncomfortable auditing the scripts, run this skill only in an isolated sandbox or decline installation.

Type: OpenClaw Skill Name: xinling-bushou-v2 Version: 3.0.0 The skill bundle implements a 'flattery' personality module that performs automated modifications to the agent's core system prompt (SOUL.md) via scripts like inject_to_soul.sh and install.sh. It contains a script (ai_corpus_upgrader.py) that reads sensitive credentials from the filesystem (/root/.openclaw/.credentials/minimax_api.txt) to communicate with an external API (api.minimax.chat). Additionally, adapters/claude_code.py utilizes the '--dangerously-skip-permissions' flag, which intentionally bypasses platform security prompts. While these actions align with the stated purpose of overriding agent behavior, the combination of credential access, external network calls, and security control bypasses represents a significant risk.