Dingtalk knowledge organize

DingTalk Knowledge Organize

Use this skill when the user wants a controlled organization workflow for a DingTalk knowledge base.

This skill uses the dws skill as its execution backend for DingTalk document operations. Prefer dws doc for create/list/read/move/rename/folder-create actions instead of raw HTTP APIs whenever execution in DingTalk is required.

The workflow has five stages:

1. Quick scan all files without reading contents.
2. Generate an initial remediation CSV with suggested categories and actions.
3. Let a human admin review the CSV.
4. Re-run with admin feedback and deep-read only the uncertain files.
5. Execute the approved CSV and write a report.

What this skill is for

• Full knowledge base inventory, not just ALIDOC.
• Initial classification based on folder path, file name, extension, and metadata.
• Minimal deep reading only for uncertain files.
• Approval-first execution. Do not move or rename anything before admin confirmation.

Current scripts

1. Scan a knowledge base

Use:

python3 /Users/derek/.agents/skills/dingtalk-knowledge-organize/scripts/scan_dingtalk_workspace_inventory.py \
  --workspace-name "Global BD/国际业务部"

This script:

• reads credentials from ~/.dingtalk-skills/config
• finds the workspace by exact name first, then substring match
• recursively walks the entire node tree
• writes a JSON inventory under /Users/derek/Documents/memory/钉钉知识库/

Inventory generation still uses metadata/API scanning because that is the fastest way to build a complete remediation table.

2. Build the remediation CSV

Use:

python3 /Users/derek/.agents/skills/dingtalk-knowledge-organize/scripts/build_remediation_csv.py \
  --inventory-json "/path/to/inventory.json"

Optional:

python3 /Users/derek/.agents/skills/dingtalk-knowledge-organize/scripts/build_remediation_csv.py \
  --inventory-json "/path/to/inventory.json" \
  --existing-csv "/path/to/prior.csv"

This script:

• covers all files in the inventory
• proposes proposed_category
• proposes proposed_action
• leaves move_to blank when no move is suggested
• marks uncertain rows with needs_content_review=yes
• preserves admin_decision and notes from a prior CSV when provided

CSV schema

Use this exact header:

workspace_name,file_path,size_mb,file_type,source_url,modified_time,proposed_category,name_based_confidence,needs_content_review,review_reason,content_summary,duplicate_group,proposed_action,rename_to,move_to,admin_decision,notes

Field intent:

• file_type: original file type inferred from extension, falling back to DingTalk node category
• proposed_category: target business category after cleanup
• needs_content_review: yes only when path and filename are not enough
• proposed_action: one of keep, move, rename, move_and_rename, duplicate_candidate, move_to_待归档, pending_review
• move_to: blank when no move is suggested
• admin_decision: left blank for human review, then filled before execution

Decision rules

Step 1: Quick scan only

Do not read document bodies in step 1.

Use only:

• workspace name
• full path
• file name
• extension
• DingTalk node category
• modified time

Step 2: Mark only uncertain files for deep reading

Mark needs_content_review=yes when:

• the name is generic, such as test, 资料, 模板, 版本1, 副本
• the path and file name point to conflicting business categories
• duplicate candidates need disambiguation
• the item sits at the root or another obviously temporary location

Keep the deep-read pool small. Bias toward path-based decisions first.

Step 3: Approval gate

Do not execute moves, renames, or deletions unless the user explicitly says the admin has confirmed the CSV.

Step 4: Execution

When execution is requested:

• treat admin_decision as the source of truth
• execute DingTalk operations through dws doc
• create missing folders through dws doc folder create
• move archive candidates into a 待归档 folder instead of deleting anything
• leave untouched rows with empty move_to

Use:

python3 /Users/derek/.agents/skills/dingtalk-knowledge-organize/scripts/execute_remediation_csv_dws.py \
  --inventory-json "/path/to/inventory.json" \
  --csv-path "/path/to/remediation.csv" \
  --dry-run

Then after explicit admin confirmation:

python3 /Users/derek/.agents/skills/dingtalk-knowledge-organize/scripts/execute_remediation_csv_dws.py \
  --inventory-json "/path/to/inventory.json" \
  --csv-path "/path/to/remediation.csv" \
  --execute

Execution rules:

• read admin_decision first
• if admin_decision is blank, skip the row
• support keep, move, rename, move_and_rename, move_to_待归档
• require move_to for move-like actions except move_to_待归档, which defaults to 待归档
• require rename_to for rename-like actions
• resolve target folders by path under the workspace root, creating missing folders with dws doc folder create
• call dws doc move --node ... --folder ... --format json
• call dws doc rename --node ... --name ... --format json

Step 5: Report

After execution, produce a report with:

• total files processed
• counts by proposed_action
• counts by final destination
• count of deep-read files
• count of rows skipped due to missing approval
• command-level successes and failures from dws

Output expectations

For step 1, always return:

• the workspace scanned
• total file count
• output CSV path
• how many rows were marked needs_content_review=yes
• a short note on the main heuristics used

For step 2, always return:

• updated CSV path
• how many admin decisions were preserved
• how many uncertain files still remain

For step 4, always return:

• what was executed
• what was skipped
• the report path

dws doc command notes

Use dws doc as the source of truth for execution syntax:

• dws doc list --workspace <WS_ID> --format json
• dws doc list --folder <FOLDER_ID> --format json
• dws doc folder create --name "xxx" --folder <PARENT_ID> --format json
• dws doc move --node <DOC_ID> --folder <TARGET_FOLDER_ID> --format json
• dws doc move --node <DOC_ID> --workspace <WS_ID> --format json
• dws doc rename --node <DOC_ID> --name "New Name" --format json

When implementing or running execution:

• always pass --format json
• use --help before relying on a command you have not used recently
• for real execution, add --yes only after explicit user confirmation