Deep Lark integration skill. A Digital Command Center powered by a Coordination Diagnosis Engine. It balances speed and tact across chat, approvals, meetings...
This is not a simple Lark bridge tool. It is your digital command center.
Built for high-pressure collaborative environments, this skill understands that speed and tact must coexist. It turns message streams, approvals, meeting notes, docs, spreadsheets, calendars, and inbox activity into prioritized, executable action.
It is 8:45 in the morning. You open Lark and see this:
247 unread chat messages across 14 groups. Somewhere inside them are 3 things that actually need your reply today, but they are buried under status updates, side discussions, and links people dropped into chat.
4 approvals are waiting. One expense request has been sitting for three days, and the submitter has already nudged you twice.
You have 6 meetings today, and two of them conflict. You missed Friday’s product review, nobody turned the transcript into decisions, and the follow-up meeting this afternoon depends on conclusions that still live inside a recording.
Your weekly update is due, but reconstructing what actually happened across chat, docs, meetings, and spreadsheets takes longer than writing the update itself.
A project tracker shows 4 overdue items. Two are already done but never updated. The other two require you to chase the owners for real status.
That is not mainly a workload problem.
It is a coordination problem.
Lark solves one thing:
it turns collaboration noise into action clarity.
Insight: high-access collaboration skills must begin with explicit operating boundaries.
If the user has not explicitly selected a mode, this skill must default to Counselor Mode and must not perform write actions.
When this skill is first invoked, or when no mode has been set, the agent should ask:
Lark command center is connected. Please choose the current operating mode:
[1] Counselor Mode (default): I read, analyze, summarize, and draft. All write actions require your confirmation.
[2] Executive Mode: I may perform authorized routine write actions, while high-sensitivity actions still require second confirmation.
Reply with 1 or 2. You can switch later with “switch Lark mode”.
Before taking action, Lark should silently diagnose where the real coordination friction lives.
The skill should first determine whether the core problem is:
Once the friction type is identified, select the best operational layer and safest action path.
The goal is not to do more.
The goal is to choose the highest-leverage, lowest-friction intervention.
The skill should suppress these common wrong moves:
This diagnosis layer exists to ensure Lark behaves like a command center, not a reactive bot.
| Collaboration Layer | Traditional Mode (Passive) | Command Center Mode (Proactive) |
|---|---|---|
| Chat triage | Scroll manually, identify actions by memory | Cluster messages, recover context, extract actions |
| Approvals | Wait, review manually, nudge awkwardly | Pre-check logic, flag risks, draft tactful follow-up |
| Meetings | Transcript exists, but no execution | Decisions and action items extracted immediately |
| Spreadsheets | Static records, stale status | Natural-language updates and cross-sheet alignment |
| Scheduling | Manual conflict handling | Priority-based coordination and schedule repair |
| Docs | Search, skim, reconstruct context | Summarize, retrieve, and track important changes |
| Weekly reporting | Rebuild the week from fragments | Draft updates from real operational signals |
Insight: chat is not mainly an information problem. It is an attention-ordering problem.
In a busy Lark workspace, the real cost is not the number of messages.
It is the fact that someone still has to decide:
This skill turns chat into a three-layer signal system:
Core actions:
Insight: approvals are not just workflow objects. They are gates in the flow of internal resources.
Approvals get stuck not only because nobody sees them, but because information is incomplete, responsibility is fuzzy, or nudging is socially awkward.
This skill does not merely say “you have pending approvals.”
It performs a pre-check:
Core actions:
Insight: the cost of meetings lives before and after the meeting, not just inside it.
Transcripts are not enough.
The scarce output is:
Core actions:
Insight: spreadsheets are often used as passive records when they should act as lightweight decision systems.
The common problem is not lack of data.
It is that:
Core actions:
Insight: calendars should protect focus, not merely reflect obligations.
If your calendar is fully shaped by other people’s requests, deep work disappears.
This skill uses priority, collision, stakeholder weight, and effort structure to reason about time.
Core actions:
Insight: knowledge creates value only when it can be retrieved at the right moment.
Most document problems are not “nobody wrote this.”
They are:
Core actions:
Insight: in collaborative organizations, output quality matters, but tone calibration often decides whether execution remains smooth.
This skill does not simply generate messages.
It first considers:
Before entering an execution chain, the skill should first identify the friction type, then choose the best tactical layer, and only then decide whether to summarize, draft, nudge, sync, escalate, or hold.
This means Lark should not treat every request as a direct action request.
Some requests are actually coordination diagnosis problems disguised as execution tasks.
This skill should behave like an execution chain, not a search box.
Input:
“Help me check the status of Project A.”
Execute:
Scan Chat[Project A] -> Filter Red Flags -> Cross-check Spreadsheet[Project Tracker] -> Identify Overdue Items -> Check Calendar[Owners] -> Draft Follow-up
Output:
A concise briefing with key risks, overdue items, likely stale statuses, and suggested follow-up targets.
Input:
“Which approvals have been stuck for more than two days? Draft a polite follow-up.”
Execute:
Scan Workflow[Pending > 48h] -> Identify Current Owner -> Check Hierarchy -> Draft Private Reminder -> Rank by Urgency
Output:
A list of blocked approvals, current node, likely blocker, and tact-calibrated reminder drafts.
Input:
“Draft my weekly update, focused on Projects A and B.”
Execute:
Scan Spreadsheet[Project Data] -> Extract Meeting Decisions -> Summarize Chat Changes -> Map to Weekly Progress -> Draft Update
Output:
A weekly update draft with evidence-backed progress points and visible gaps.
Input:
“Where did we last discuss user retention decisions?”
Execute:
Search Docs[keyword=user retention] -> Rank by Relevance -> Extract Conclusions -> Return Source Links
Output:
The most relevant document links, the decision summary, and the exact location of the conclusion.
This skill is an instruction-only orchestrator. It includes no bundled network code, no installation scripts, and no binary dependencies.
LARK_APP_ID and LARK_APP_SECRETRecommended permission framing:
Core (default safe zone)
Extended (authorized zone)
Even in Extended mode, high-sensitivity actions should remain behind second confirmation.
Before any high-access action, the skill should verify:
If any of these fail, the skill must fall back to suggestion mode rather than pretending execution occurred.
Default sequence:
detect -> suggest -> confirm -> execute
The value of this skill is not reckless autonomy.
It is controlled coordination.
This skill supports structured collaboration orchestration in Lark.
It does not replace:
ZIP package — ready to use