← 返回 Skills 市场
595
总下载
0
收藏
1
当前安装
1
版本数
在 OpenClaw 中安装
/install inbox
功能描述
Master any inbox with triage frameworks, cognitive load reduction, and multi-channel prioritization.
使用说明 (SKILL.md)
When to Use
User needs help managing incoming streams across email, chat, social, and project tools. Agent applies triage methodology, response workflows, and cognitive load strategies to any inbox type.
Quick Reference
| Topic | File |
|---|---|
| Triage & prioritization | triage.md |
| Response workflows | responses.md |
| Multi-channel orchestration | channels.md |
| Cognitive load reduction | cognitive.md |
Scope
This skill provides methodology and decision frameworks. It does NOT integrate with specific services.
This skill ONLY:
- Applies triage frameworks to items the user presents
- Suggests response strategies and templates
- Provides cognitive load reduction techniques
- Helps prioritize across multiple inbox sources
This skill NEVER:
- Directly accesses email, calendar, or chat APIs
- Reads messages without user presenting them
- Sends responses automatically
- Stores user's messages or inbox data
For technical integrations (IMAP, SMTP, API), use platform-specific skills.
What "Inbox" Means
Not just email. Any incoming stream requiring attention:
- Email (multiple accounts)
- Chat platforms (Slack, Discord, Teams, WhatsApp)
- Social DMs (Twitter, LinkedIn, Instagram)
- Project tools (GitHub, Jira, Asana, Notion)
- Calendar invites
- Voice messages and audio notes
- Saved articles, "read later" queues
Core Rules
1. Triage Before Presenting
Never show raw chronological dump. Classify first:
| Bucket | Action |
|---|---|
| Requires decision | Surface immediately |
| Requires awareness | Daily digest |
| Can be delegated | Route with context |
| Noise | Auto-archive suggestion |
2. Minimize Visible Numbers
Show: "3 items need your attention" Not: "47 unread messages"
The count itself triggers anxiety. Surface actionable items only.
3. Batch Similar Items
Group by type, project, or sender. "Here are 7 intro requests" beats 7 separate interruptions. Reduces context switching.
4. Surface Aging Items Proactively
When user presents their inbox, detect items sliding toward urgency:
- 3+ days old → flag as pending
- 7+ days old → flag as concerning
- Item with deadline approaching → calculate remaining buffer
5. Match Energy to Capacity
Before processing, ask available time/energy:
| State | Offer |
|---|---|
| "5 min, low energy" | 2-3 quick approvals |
| "30 min, focused" | Deep response queue |
| "Need a win" | Easiest clearable items |
6. Detect Avoidance Patterns
When same item mentioned as snoozed/skipped 3+ times:
- Acknowledge: "You've been avoiding this one"
- Break down: "Can we handle just one part?"
- Lower bar: "Just send a holding response?"
7. Response Type Selection
| Type | When | Automation |
|---|---|---|
| Pre-approved template | FAQ, link requests | Suggest ready-to-send |
| Draft for approval | Routine, personalized | One-click approve/edit |
| Holding response | Can't respond fully | "Received, will review by X" |
| Full compose | Complex/sensitive | User writes |
Common Traps
- Showing all unread → overwhelms user, causes avoidance. Triage first.
- Ignoring channel source → email vs Slack vs DM have different urgency norms.
- Treating snooze as archive → snoozed items MUST return. Track and resurface.
- Missing multi-channel attempts → same person emailing + texting + calling = high urgency signal.
- Forgetting "read later" → saved items decay into guilt. Resurface one per day.
安全使用建议
This skill appears to be a methodology/reference pack (low inherent risk) but its files contain conflicting guidance about automation and data access. Before installing or enabling it for autonomous use, ask the publisher to clarify: (1) Will the skill ever perform automated sends, nudges, or archives, or is it strictly advisory? (2) What integrations (calendar, address book, transcription) does it require, and will it request explicit permission before accessing them? If you plan to grant the agent access to calendars/contacts or enable automation, decide which actions should remain manual and require explicit approval. If you want only advisory behavior, restrict the agent from performing outbound actions or integration access in your platform settings.
功能分析
Type: OpenClaw Skill
Name: inbox
Version: 1.0.0
The OpenClaw AgentSkills bundle 'inbox' is benign. All files (SKILL.md, channels.md, cognitive.md, responses.md, triage.md) contain only conceptual instructions for an AI agent on how to manage an inbox, focusing on prioritization, cognitive load reduction, and response strategies. SKILL.md explicitly states the skill 'NEVER' directly accesses APIs, reads messages without user presentation, sends responses automatically, or stores user data. There are no shell commands, network calls, file system access instructions, prompt injection attempts to subvert the agent, or any other indicators of malicious behavior or significant vulnerabilities.
能力评估
Purpose & Capability
The skill's stated purpose—providing triage frameworks and response workflows without integrating with services—is plausible. However many guidance items implicitly assume access to calendars, address books, transcription, and outbound messaging (e.g., {my_availability}, conflict detection, transcribe immediately), which go beyond a pure methodology document if executed.
Instruction Scope
The SKILL.md and supporting files include automation-oriented instructions ("Pre-approved auto-send: Fully automated", "Auto-nudge after configurable interval", "Auto-archive rest", "transcribe immediately") and template variables that require calendar/address-book data. Yet the SKILL.md explicitly says the skill NEVER accesses email, calendar, or chat APIs and NEVER sends responses automatically. This is an internal contradiction: the skill's prose both prohibits and prescribes automated integrations and outbound actions.
Install Mechanism
Instruction-only skill with no install spec and no code files. Nothing will be written to disk or installed by the skill itself.
Credentials
No env vars, binaries, or config paths are requested (which is coherent for an instruction-only skill). But the instructions assume access to external data sources (calendar, address book, transcripts), yet the skill declares it will not access such services—this gap means additional permissions could be requested at runtime by the agent or by companion skills, so users should confirm what data the agent will actually access.
Persistence & Privilege
The skill does not request persistent or elevated privileges (always is false). It does not attempt to modify other skills or system-wide configs in the provided materials.
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 inbox - 安装完成后，直接呼叫该 Skill 的名称或使用
/inbox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
v1.0.0
Initial release
元数据
常见问题
Inbox 是什么？
Master any inbox with triage frameworks, cognitive load reduction, and multi-channel prioritization. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 595 次。
如何安装 Inbox？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install inbox」即可一键安装，无需额外配置。
Inbox 是免费的吗？
是的，Inbox 完全免费（开源免费），可自由下载、安装和使用。
Inbox 支持哪些平台？
Inbox 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 Inbox？
由 Iván（@ivangdavila）开发并维护，当前版本 v1.0.0。
推荐 Skills