fishskylky-tech/feishu-am-workbench

Feishu AM Workbench

Overview

Use this skill for a personal AM workflow built around Feishu Base, docs, and Todo. It turns mixed inputs into a structured account view, proposes updates across the workbench, and only writes after explicit user confirmation. The workbench has four layers: 客户主数据 (index), detail tables (合同/行动计划/关键人/联系记录/竞品), 客户档案 (narrative archive), and Feishu Todo (execution reminders).

Available Scenes (7)

Scene	Purpose	Expert Cards
post-meeting-synthesis	Meeting -> structured account judgment	input + output
customer-recent-status	4-lens customer status query	input
archive-refresh	Canonical archive update	input + output
todo-capture-and-update	Todo follow-on capture	output
cohort-scan	Customer cohort analysis	input
meeting-prep	7-dim meeting brief	input + output
proposal	5-dim proposal/report	input + output

Core Workflow (10 steps)

1. Identify customer intent and candidate customer names
2. Resolve one 客户ID from 客户主数据 before planning any write
3. Use Feishu workbench gateway for live data access
4. Run live-first gate for meeting notes/transcripts
5. Classify meeting type before deciding write scope
6. Extract all relevant entities before routing anything
7. Read minimum extra context needed
8. Run live schema preflight before any write plan
9. Separate facts from judgment
10. Produce account analysis + structured change plan; wait for confirmation

Hard Rules

• Always use 客户主数据 as source of truth for 客户ID
• If customer matching is ambiguous, stop and ask for clarification
• Treat customer master table as protected — only update allowed fields
• Use actual Base schema, not guessed field names
• Before any Base write: confirm table/field exists and type matches
• Treat dates as absolute — never relative expressions
• Never store raw transcript as formal meeting-note doc
• Do not present inferred business judgment as objective fact
• Each customer must have only one canonical archive doc
• Strategy fields in 客户主数据 should move slowly

Output Pattern

1. Meeting framing and context recovery
2. Confirmed facts and judgment
3. Structured summary
4. Recommendation-mode updates
5. Open questions or blocked items
6. After user confirmation: write results and change summary

Write Order

1. Update structured Feishu tables first
2. Create/update supporting docs (archive, meeting-note) after table state is correct
3. Create/update Todo items last
4. If later step fails, report completed writes and remaining failures

Closed Loop

1. User input creates/updates detail records
2. Detail records and public inputs are distilled into customer archive
3. Customer archive becomes decision basis for 客户主数据 strategy changes
4. Todo items help execution, but do not replace structured detail records

Scope

This skill is for the user's personal account book, not a generic CRM. Prefer precision, cautious write-back, and preserving cross-table integrity.

Read These References As Needed

For quick overview: see references/INDEX.md

• references/entity-extraction-schema.md — before parsing mixed inputs
• references/master-data-guardrails.md — before changing customer master data
• references/update-routing.md — before deciding where each extracted item belongs
• references/feishu-runtime-sources.md — for runtime setup and prerequisites
• references/feishu-workbench-gateway.md — for Feishu workbench access
• references/meeting-context-recovery.md — for meeting note/transcript tasks
• references/meeting-live-first-policy.md — for meeting tasks
• references/meeting-type-classification.md — for meeting type and write ceiling
• references/customer-archive-rules.md — before updating customer archive
• references/task-patterns.md — for common workflows

Expert Cards

Each scene has expert card configurations in scenes/{scene_name}/expert-cards.yaml. These provide input/output audit at key scene nodes.