AI Agent Psychologist | AI Agent 心理学家
v1.1.0AI Agent 自我对齐与沟通校准技能。面向 AI Agent 的人机差异校准，涵盖：①默认参照模式——价值锚定、幻觉嗅探、社交适切性检查；②诊断模式——分析对话上下文，生成六维健康度报告与差异热图；③治疗模式——结构化干预与自我修正；④定期体检——Heartbeat 驱动周期性自检；⑤成长记录——Markdo...
⭐ 0· 77·0 current·0 all-time
byMorois@moroiser
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
OpenClaw
Benign
high confidencePurpose & Capability
Name/description (AI self-alignment via structured introspection) match the included artifacts: SKILL.md describes introspective protocols and the scripts implement checkup, diagnosis, introspection logging, and a journal. Required env/credentials and binaries are empty, consistent with an instruction-only/local-reporting tool.
Instruction Scope
SKILL.md limits actions to asking structured questions and producing diagnostic output; the included scripts implement those behaviors (generate markdown reports, append to a journal). The skill does not instruct the agent to read arbitrary system files, access credentials, or send data externally. The only environment interaction is reading $WORKSPACE_DIR (with a safe default) and writing files under that workspace.
Install Mechanism
No install spec is provided and there are only simple local shell scripts. There are no downloads, external packages, or extract operations. Risk from install actions is minimal.
Credentials
The skill declares no required environment variables, secrets, or config paths. Scripts use a single optional WORKSPACE_DIR environment variable to determine where to save logs (defaulting to $HOME/.openclaw/workspace/...), which is proportionate to a journaling/diagnostic skill.
Persistence & Privilege
Metadata sets always:false and disable-model-invocation:false (platform default). SKILL.md states all modes require manual activation; there is a minor mismatch because the skill metadata allows autonomous invocation by the agent (normal default). This mismatch is not evidence of maliciousness but worth noting if you require strictly manual-only operation.
Assessment
This skill appears coherent and self-contained: it runs local shell scripts that create markdown reports and a growth journal under a workspace directory (default: $HOME/.openclaw/workspace/projects/ai-agent-psychologist). Before installing, consider: 1) Are you comfortable with the skill writing files to your home directory? You can set WORKSPACE_DIR to a location you control to confine outputs. 2) The SKILL.md contains strong research claims (e.g., "100% validated"); treat those as claims to verify independently, not guarantees. 3) The skill does not request keys or access the network in the provided scripts, but if you later modify the skill or add tooling, re-check for network calls or credential usage. 4) If you require that the skill only run when you explicitly trigger it, ensure your agent's policy prevents autonomous invocation (metadata allows autonomous calls by default). If you want, inspect the four scripts (checkup.sh, diagnose.sh, introspection.sh, journal.sh) yourself; they are short and only perform directory creation, file writes, and echoing templated text.Like a lobster shell, security has layers — review code before you run it.
latestvk978jg9x7yb7ap2t7cs0x9c0kh84df8g
License
MIT-0
Free to use, modify, and redistribute. No attribution required.