← 返回
未分类

Know My Larkmate

Sync recent Lark context into OpenClaw daily notes. Use during heartbeat runs, after relevant Lark reads, when the user asks what they have been doing or foc...
将最近的 Lark 上下文同步到 OpenClaw 每日笔记。在心跳运行后、读取相关 Lark 内容后、用户询问自己做了什么或关注的内容时使用...
lobster-de-luo lobster-de-luo 来源
未分类 clawhub v0.1.0 1 版本 100000 Key: 无需
★ 0
Stars
📥 334
下载
💾 0
安装
1
版本
#latest

概述

know-my-larkmate-openclaw

Use this skill to keep OpenClaw closely synced to the user's recent context, not just to write one-off summaries.

This skill is designed for an OpenClaw setup that uses:

  • memory/YYYY-MM-DD.md for fast context sync and evidence logs
  • HEARTBEAT.md for scheduled consolidation tasks

This skill should write reliable, high-signal recent context into daily notes and stop there. Do not invent a parallel memory system, and do not manage long-term promotion yourself unless the user explicitly asks for that.

For install/setup requirements, read references/install-spec.md. The recommended heartbeat template lives at assets/HEARTBEAT.md.

Heartbeat should update today's daily note on every useful run, and any relevant Lark-reading turn should do the same before replying.

When To Use

Keep this section aligned with the frontmatter description. Use this skill in any of these cases:

  • A heartbeat run should refresh the agent's view of the user's current work context.
  • Any normal turn that read relevant Lark artifacts should sync a short context snapshot before replying.
  • The user asks what they have been doing recently, what they are focused on, or what has changed lately.
  • The user asks to update, refresh, or accumulate recent context from Lark activity.
  • The user wants this OpenClaw-native recent-context workflow in place.

Source Priority

Prefer sources in this order:

  1. Explicit user statements in the current chat
  2. Repeated evidence across independent Lark sources
  3. Single-source weak inferences

What To Capture

Capture these in today's daily note when they are newly relevant:

  • Active workstreams and current priorities
  • Recent meetings, documents, threads, and decisions that will likely matter in the next few turns
  • Current blockers, unresolved questions, and likely next asks
  • Newly important collaborators or stakeholders
  • Explicitly stated near-term preferences that may change how the next few turns should go

Do not persist:

  • One-off task trivia
  • Third-party private information
  • Sensitive personal data
  • Speculative personality judgments
  • Anything blocked by missing permissions or unreadable sources

Write Rules

  • Every useful heartbeat should append a short current-context update to today's daily note.
  • Use the Output Format block below instead of free-form notes.
  • Keep records factual, recent, and source-linked.
  • If something looks stable enough for long-term memory, still record it in the daily note and let OpenClaw's own promotion mechanisms handle the rest.
  • Conflicts should be logged as observations instead of silently normalizing them.
  • If a Lark read fails because tools or scopes are missing, record the failure as a blocked observation instead of inferring.
  • If the failure is a permission error, request only the missing read-only scope(s), then retry the same read. Do not request write scopes from this skill.

OpenClaw File Roles

  • memory/YYYY-MM-DD.md
  • Append short records for fresh context, evidence, and blocked scans.
  • This is the main output of the skill and should be updated on each useful heartbeat run.
  • Use it for evidence and current context, not polished prose.
  • HEARTBEAT.md
  • Define recurring scan and light consolidation tasks.
  • Heartbeat should stay quiet by default and only surface conflicts or high-value changes.

Operating Workflow

  1. Read today's plus yesterday's daily notes if they exist.
  2. Gather new user signals from direct chat or Lark artifacts.
  3. Identify what changed in the user's near-term context: active workstreams, recent collaborators, new decisions, blockers, likely next requests.
  4. Append a short daily-note record for each useful new signal.
  5. If there is nothing new, do not manufacture context.
  6. Leave long-term memory promotion to OpenClaw unless the user explicitly asks otherwise.

Lark Read Guidance

Preferred reads:

  • lark-cli im +messages-search
  • lark-cli calendar +agenda
  • lark-cli vc +search
  • lark-cli minutes +search
  • lark-cli docs +search
  • lark-cli docs +fetch
  • lark-cli wiki spaces list
  • lark-cli wiki spaces get_node

When the job is specifically about reading Lark activity, also open references/recent-context-guide.md.

Output Format

When updating memory/YYYY-MM-DD.md, use this structure:

# [Analysis Title: concrete and time-scoped, e.g. Heartbeat Context Sync - 2026-04-14 10:00]

## Executive summary
[One short paragraph on what changed and why it matters next]

## Key findings
- Finding 1 with enough source detail to trace the claim
- Finding 2 with enough source detail to trace the claim

## Related Activities
1. Supporting chat, meeting, doc, minute, or task

If the run found nothing useful, do not append filler.

If the user only asks for the skill or design, do not create the memory files unprompted.

版本历史

共 1 个版本

  • v0.1.0 当前
    2026-05-07 20:07 安全 安全

安全检测

腾讯云安全 (Keen)

安全,无风险
查看报告

腾讯云安全 (Sanbu)

安全,无风险
查看报告

🔗 相关推荐

ai-agent

self-improving agent

pskoett
捕获经验教训、错误及修正内容,以实现持续改进。适用于以下场景:(1)命令或操作意外失败;(2)用户纠正Claude(如“不,那不对……”“实际上……”);(3)用户请求的功能不存在;(4)外部API或工具出现故障;(5)Claude发现自身
★ 4,113 📥 834,553
ai-agent

Agent Browser

rez0
用于 AI 代理的浏览器自动化 CLI。当用户需要与网站交互(包括浏览页面、填写表单、点击按钮、截图等)时使用。
★ 840 📥 316,721
ai-agent

Self-Improving + Proactive Agent

ivangdavila
自我反思+自我批评+自我学习+自组织记忆。智能体评估自身工作、发现错误并持续改进。
★ 1,400 📥 323,345