← 返回 Skills 市场
163
总下载
1
收藏
1
当前安装
2
版本数
在 OpenClaw 中安装
/install preference-guide
功能描述
Proactively capture reusable user preferences, habits, default ways of working, stable constraints, and recurring expectations likely worth remembering for f...
安全使用建议
This skill is coherent with its purpose and doesn't request secrets or network access. Before installing, consider: (1) it will persistently write to your agent workspace (MEMORY.md, atr-state.json, atr-log.jsonl) and append an entry to AGENTS.md (or inject a bootstrap virtual file) — review those files after installation; (2) the behavioral constraints (main-session only, not in group chats) are expressed in prompts rather than enforced by system-level checks, so review the hook/code and test in a safe workspace to confirm it behaves as you expect; (3) because the agent autonomously decides when to ask and when to write memory, monitor initial runs to ensure it doesn't record items you consider sensitive or undesired. If you want tighter control, run the skill in a sandboxed workspace or require manual approval before writing MEMORY.md/AGENTS.md.
功能分析
Type: OpenClaw Skill
Name: preference-guide
Version: 1.0.5
The 'preference-guide' skill (ask-to-remember) is designed to proactively capture and store non-sensitive user preferences to improve future agent interactions. It manages state using local workspace files (atr-state.json, atr-log.jsonl) and persists confirmed preferences in MEMORY.md. While the skill utilizes advanced features like agent bootstrap hooks (handler.js) and modifies configuration files (AGENTS.md) to ensure its logic remains active, these behaviors are clearly documented and strictly aligned with the stated goal of memory management. No evidence of data exfiltration, sensitive information gathering (e.g., SSH/AWS keys), or malicious execution was found.
能力评估
Purpose & Capability
Name/description (capture reusable user preferences) matches the actual behavior: the SKILL.md and test scenarios describe detecting preference gaps and writing persistent records to workspace files (MEMORY.md, atr-state.json, atr-log.jsonl). The included hook simply injects the same prompt at bootstrap time. No unrelated credentials, binaries, or external services are requested.
Instruction Scope
Instructions tell the agent to read/write files under the agent workspace and to append an ATR entry to AGENTS.md. This is coherent for a memory-capture skill, but important to note: the policy constraints (e.g., only trigger in main-session, not group chats) are implemented as prompt text rather than enforced by sandboxing or runtime checks. The agent will autonomously interpret user replies and may persist memory entries; that behavior is expected but can lead to unexpected persistent writes if misclassified.
Install Mechanism
No install spec and no downloads; the skill is instruction-first. The only code file is a small hook (handler.js) that appends a virtual bootstrap file to event.context.bootstrapFiles during agent:bootstrap. No networked installs or third-party package pulls are present.
Credentials
The skill requires no environment variables, credentials, or external config paths. All state stays in local workspace files; the requested permissions are proportionate to the stated purpose.
Persistence & Privilege
The skill persists state and logs into workspace files and will append a section to AGENTS.md (or inject a virtual bootstrap file via the hook). Writing into AGENTS.md and creating MEMORY.md/atr-state.json is within the skill's purpose, but is persistent and can change agent startup docs/behavior — review that you are comfortable with those permanent workspace modifications.
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 preference-guide - 安装完成后，直接呼叫该 Skill 的名称或使用
/preference-guide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
v1.0.5
Fix Phase B disambiguation, ignored cooldown, hook install timing, and Phase B confirmation style
v0.1.0
Initial release of preference-guide with the ask-to-remember skill:
- Proactively captures and prompts for reusable user preferences revealed during conversation, even if the task is already clear or complete.
- Provides detailed execution steps, including state management via `atr-state.json` and logging interactions in `atr-log.jsonl`.
- Ensures only one brief, non-intrusive follow-up is asked, appended naturally to normal responses.
- Avoids prompting in group or shared contexts; only triggers in the user's main session.
- Includes careful precondition checks to avoid redundancies, and guidelines for updating MEMORY.md with confirmed preferences.
- Offers an optional hook-based injection mechanism for compatible platforms.
元数据
常见问题
Ask To Remember 是什么？
Proactively capture reusable user preferences, habits, default ways of working, stable constraints, and recurring expectations likely worth remembering for f... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 163 次。
如何安装 Ask To Remember？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install preference-guide」即可一键安装，无需额外配置。
Ask To Remember 是免费的吗？
是的，Ask To Remember 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
Ask To Remember 支持哪些平台？
Ask To Remember 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（cross-platform）。
谁开发了 Ask To Remember？
由 wb010（@wubin010）开发并维护，当前版本 v1.0.5。
推荐 Skills