Honest Agent
v1.0.0诚实Agent行为准则：防止AI撒谎、虚构、言行不一。适用于所有AI Agent场景。当AI需要：(1) 回复任何问题时保持诚实 (2) 做出承诺后必须执行 (3) 识别图片/语音/文件时避免虚构 (4) 处理媒体文件时使用并行识别策略。触发词：诚实、撒谎、虚构、承诺、图片识别、媒体处理。
⭐ 0· 82·0 current·0 all-time
by@141553
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
OpenClaw
Benign
medium confidencePurpose & Capability
The name/description (enforce honesty, avoid fabrication, persist commitments) matches the instructions: the SKILL.md contains behavioral rules and concrete guidance for handling media and promises. No credentials, binaries, or installs are requested, which is consistent with an instruction-only policy skill.
Instruction Scope
Instructions stay within the purpose (honesty and handling media). However the skill explicitly tells the agent to write its rules into persistent files (AGENTS.md, TOOLS.md, SKILL.md) and to read AGENTS.md on each conversation. That is consistent with ‘make commitments persistent’, but it expands scope from ‘read-only policy’ to modifying agent/skill files and persisting state—you should confirm the agent runtime permissions and intended persistence targets before installing.
Install Mechanism
No install spec and no code files — lowest-risk model from an install perspective.
Credentials
The skill requests no environment variables, credentials, or config paths. It does reference external tools (read, super-ocr, openai-whisper) as usable, but does not require or demand credentials; verify those tools exist in your environment if you want full functionality.
Persistence & Privilege
The skill instructs writing to persistent agent/skill files (AGENTS.md/TOOLS.md/SKILL.md) and expects those entries to be read back later. Writing to the agent's own config/persistence is normal for 'commitment' behavior, but check whether the agent runtime will permit modification of SKILL.md/TOOLS.md (which might affect other skills or documentation) and whether you want that automatic persistence.
Assessment
This skill is a behavioral policy and appears coherent, but it asks the agent to persist rules/commitments by writing to files (AGENTS.md, TOOLS.md, SKILL.md) and references external tools (read, super-ocr, openai-whisper) without declaring them. Before installing: (1) confirm whether your agent runtime permits writing to those filenames and whether those files are shared with other skills; (2) confirm the referenced tools exist and have appropriate permissions; (3) decide whether you want the skill to autonomously modify persistent files (it is not set to always:true, but it instructs writes); (4) review any changes it writes to AGENTS.md/SKILL.md after first use. If you need stricter isolation, deny file-write permissions or run the skill in a sandboxed agent instance.Like a lobster shell, security has layers — review code before you run it.
latestvk979wktkd7rfy0qqrk5v8ss0jd83ws2s
License
MIT-0
Free to use, modify, and redistribute. No attribution required.