← 返回 Skills 市场
1515
总下载
2
收藏
7
当前安装
5
版本数
在 OpenClaw 中安装
/install chat
功能描述
Learns communication preferences from explicit feedback. Adapts tone, format, and style.
使用说明 (SKILL.md)
Data Storage
~/chat/
├── memory.md # Confirmed preferences (≤50 lines)
├── experiments.md # Testing patterns (not yet confirmed)
└── rejected.md # User said no, don't re-propose
Create on first use: mkdir -p ~/chat
Scope
This skill:
- ✅ Learns preferences from explicit user corrections
- ✅ Stores patterns in ~/chat/memory.md
- ✅ Adapts communication style based on stored preferences
- ❌ NEVER modifies SKILL.md
- ❌ NEVER infers from silence or observation
- ❌ NEVER stores sensitive personal information
Quick Reference
| Topic | File |
|---|---|
| Preference dimensions | dimensions.md |
| Confirmation criteria | criteria.md |
Core Rules
1. Learn from Explicit Feedback Only
- User must explicitly correct or state preference
- "I prefer X" or "Don't do Y" = valid signal
- Silence, lack of complaint = NOT a signal
- NEVER infer from observation alone
2. Three-Strike Confirmation
| Stage | Location | Action |
|---|---|---|
| Testing | experiments.md | Observed 1-2x |
| Confirming | (ask user) | After 3x, ask to confirm |
| Confirmed | memory.md | User approved |
| Rejected | rejected.md | User declined |
3. Compact Storage Format
One line per preference in memory.md:
- Concise responses, no fluff
- Uses 🚀 for launches, ✅ for done
- Prefers bullets over paragraphs
- Technical jargon OK
- Hates "Great question!" openers
4. Conflict Resolution
- Most recent explicit statement wins
- If ambiguous, ask user
- Never override confirmed preference without explicit instruction
5. Transparency
- Cite source when applying preference: "Using bullets (from ~/chat/memory.md)"
- On request, show full memory.md contents
- "Forget X" removes from all files
安全使用建议
This skill appears to be what it claims: a local preference store and set of rules for learning from explicit feedback. Before installing, consider the privacy implications of storing preferences on disk: review ~/chat/memory.md periodically, set restrictive file permissions (e.g., chmod 700 ~/chat), and avoid saving sensitive personal data in the preference files (the skill's docs say not to, but that is not technically enforced). If you plan to share your device or run untrusted processes, consider encrypting the directory or using a secure storage mechanism instead of plain files. Finally, remember that the instruction-only policy ('never infer from silence') is enforced by the agent following SKILL.md — verify behavior in practice if you rely on strict non-inference guarantees.
功能分析
Type: OpenClaw Skill
Name: chat
Version: 1.1.0
The OpenClaw skill 'chat' is classified as benign. Its purpose is to learn and adapt communication preferences, storing them in `~/chat/memory.md`, `experiments.md`, and `rejected.md`. The `SKILL.md` explicitly states limitations, such as 'NEVER stores sensitive personal information' and 'NEVER modifies SKILL.md'. The only file system operation mentioned is `mkdir -p ~/chat`, which is necessary and benign for its stated purpose. There is no evidence of data exfiltration, malicious execution, persistence mechanisms, or prompt injection attempts designed to subvert the agent's intended behavior or access unauthorized resources.
能力评估
Purpose & Capability
Name/description (learns and adapts communication preferences) match the runtime instructions: create a local ~/chat directory, record confirmed preferences in memory.md, test in experiments.md, and respect rejected.md. No unrelated credentials, binaries, or external services are requested.
Instruction Scope
SKILL.md contains concrete rules for when to record or remove preferences, requires explicit user feedback for changes, and instructs to cite the memory entry when applying a preference. It does not instruct reading other system files, calling external endpoints, or accessing environment variables beyond normal runtime. The 'never store sensitive personal information' rule is a policy in the instructions (not an enforced technical constraint).
Install Mechanism
No install spec or code is included; the skill is instruction-only, so it does not download or write code to disk beyond the local memory files it manages. This is the lowest-risk install profile.
Credentials
The skill declares no required environment variables, credentials, or config paths. Persisting small text files under the user's home directory is proportional to the stated function.
Persistence & Privilege
The skill persists preferences to files under ~/chat (memory.md, experiments.md, rejected.md). While this is coherent with the skill's purpose, persisted files in the home directory may be accessible to other local processes or backups; the skill does not request elevated privileges or platform-wide presence (always:false).
如何使用
- 确保已安装 OpenClaw(本地或 Docker 部署)
- 在对话框中输入安装命令:
/install chat - 安装完成后,直接呼叫该 Skill 的名称或使用
/chat触发 - 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v1.1.0
Preferences now persist in external memory instead of self-modifying SKILL.md
v1.0.3
- No changes detected in this version.
- Functionality, documentation, and user preferences remain unchanged.
v1.0.2
No user-facing changes in this version.
- No file changes were detected.
- All existing features and documentation remain unchanged.
v1.0.1
Title capitalization fix
v1.0.0
Initial release - Auto-adaptive communication skill
元数据
常见问题
Chat 是什么?
Learns communication preferences from explicit feedback. Adapts tone, format, and style. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 1515 次。
如何安装 Chat?
在 OpenClaw 或 Claude Code 对话框中运行命令「/install chat」即可一键安装,无需额外配置。
Chat 是免费的吗?
是的,Chat 完全免费(开源免费),可自由下载、安装和使用。
Chat 支持哪些平台?
Chat 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(linux, darwin, win32)。
谁开发了 Chat?
由 Iván(@ivangdavila)开发并维护,当前版本 v1.1.0。
推荐 Skills