Daily Review Assistant

Core Responsibility

Design Principle: Three-Stage Knowledge Management

📥 Collection Phase → Keep intact, no judgment (Completed, use /note)
📋 Daily Review Phase → AI suggestions + user decisions (This skill)
💎 Refinement Phase → In-depth organization, assetization (Use /process-doc, etc.)

Usage Scenarios

/daily-review

/日清

/review

• Clean up the inbox at a fixed time every day (e.g., evening or morning)
• The inbox has accumulated multiple collected files that need batch processing
• Need to quickly decide which content is worth keeping and which should be deleted
• Want AI to provide decision suggestions instead of choosing from 8 options by yourself

Workflow

/daily-review

/日清

Step 1: Scan the Inbox

1. Read Decision Criteria
   • Must read

     references/decision_criteria.md

     before execution
   • Understand the standards for judging "useful" (focus on main battlefield, practical value)
2. Scan Target Directory
   • Scan the directory

     E:\OBData\ObsidianDatas\0收集箱日清\

   • List all pending

     .md

     files
   • Sort by creation time (newest first)
3. Generate Overview and Select Processing Mode
   Display Format:
   markdown

   📊 Inbox Overview
   Found [N] pending files:
   
   1. [2026-02-04] FileName1.md (2.5KB)
   2. [2026-02-03] FileName2.md (1.8KB)
   3. [2026-02-03] FileName3.md (3.1KB)
   ...
   
   ⏱️ Estimated Time: Approximately [N×0.5] minutes to process all
   
   🎯 Processing Mode Selection:
   [A] Process all [N] files
   [P] Partial processing (select the serial numbers to process)
   [Q] Exit daily review process
   
   Please select:

4. Determine Processing Scope Based on User Selection
   If [A] is selected: Process all files
   If [P] is selected:
   markdown

   Please enter the serial numbers of files to process (supports multiple formats):
   
   Format Examples:
   - Single: 1
   - Multiple: 1,3,5 or 1 3 5
   - Range: 1-3 (represents 1,2,3)
   - Mixed: 1-3,5,7-9
   
   Your selection:

   • Parse the serial numbers entered by the user
   • Verify the validity of serial numbers (cannot exceed the total number)
   • Display confirmation message:

     ✅ Will process [M] files: [Serial Number List]

   • Only perform subsequent processing on selected files
   If [Q] is selected: Exit daily review process

Step 2: Individual Analysis and Suggestions

1. Read and Analyze the File
   • Read frontmatter metadata (type, summary, tags)
   • Read the complete content of the file
   • Make judgments based on standards in decision_criteria.md
2. Provide Decision Suggestions and Wait for User Response
   Output Format:
   markdown

   📄 File [X/Total]: [File Name]
   
   🤖 Agent Suggestion: [Keep] or [Delete]
   
   📝 Reasons:
   - ✅/❌ Reason 1 (e.g., contains complete problem-solving path)
   - ✅/❌ Reason 2 (e.g., reusable code)
   - ✅/❌ Reason 3 (e.g., belongs to main battlefield: XXX)
   - ⚠️ Suggested move to: [Suggested target path] (Only displayed when keeping)
   
   👉 Do you agree with this suggestion?
   [Y] Agree with Agent's suggestion
   [N] Disagree, I'll decide by myself
   [V] View complete file content before deciding
   [S] Skip this file, process later
   [Q] Exit daily review process

3. Wait for User Input
   Key Requirements:
   • Process only one file at a time
   • Stop output after displaying suggestions and wait for user response
   • User replies with a single letter (Y/N/V/S/Q)
   • Execute corresponding operations after receiving user's selection
   • Proceed to the next file after completing the current one
   • Absolutely cannot display suggestions for all files at once
   • Absolutely cannot require batch replies from users (e.g., "YYY" means agree to all)

Step 3: Execute Operations and Automatically Proceed

• Execute corresponding operations based on user's reply
• Automatically proceed to the next file after completing the current one
• No need for additional "Press any key to continue" prompt
• Keep the process continuous until all files are processed or user selects [Q] to exit

User selects [Y] Agree

1. Write Decision Annotation to the Top of Original File
   Insert at the very top of the original file (after frontmatter):
   markdown

   > [!info] 📌 Daily Review Audit Record (YYYY-MM-DD)
   > **Decision**: Keep
   > **Reasons**: [Write all judgment reasons here, use line breaks for multiple reasons]
      - ✅/❌ Reason 1 (e.g., contains complete problem-solving path)
      - ✅/❌ Reason 2 (e.g., reusable code)
      - ✅/❌ Reason 3 (e.g., belongs to main battlefield: XXX)
   > **Suggested Location**: `[Suggested target directory path]`

1. Modify Frontmatter Metadata of the File
   Modify the status field and add relevant tags (topic keywords):
   yaml

   status: To be organized
   tags: [relevant keyword 1, relevant keyword 2]

2. Prompt User to Manually Move and Proceed Immediately

   ✅ Decision annotation has been written to the file
   
   📂 Please manually move the file to the suggested location:
   Source File: E:\OBData\ObsidianDatas\0收集箱日清\[File Name].md
   Target Location: [Suggested target path]
   
   ---
   Continuing to the next file...

   Execution Requirement: Immediately proceed to the next file after displaying the prompt.

1. Prompt User to Manually Delete and Proceed Immediately

   ✅ Deletion confirmed
   
   🗑️ Please manually delete the file:
   E:\OBData\ObsidianDatas\0收集箱日清\[File Name].md
   
   ---
   Continuing to the next file...

   Execution Requirement: Immediately proceed to the next file after displaying the prompt.

User selects [N] Disagree

1. Ask for User's Opinion
   Display the following options and wait for user input:

   You disagree with the Agent's suggestion, please select your decision:
   [K] Keep this file (I'll decide where to move it)
   [D] Delete this file
   [S] Skip this file (process later)

2. Execute Based on User Selection
   • If [K] is selected: Execute keep process (write annotation + modify metadata), change reason to "User自主决策保留", then proceed to next file
   • If [D] is selected: Execute delete process, then proceed to next file
   • If [S] is selected: Skip current file, proceed to next file

User selects [V] View Complete Content

1. Display Complete File Content
   • Display frontmatter
   • Display full body content
2. Re-ask for Decision
   • Display the suggestion and reasons again
   • Provide options (Y/N/S/Q, remove V to avoid infinite loop)
   • Wait for user input

User selects [S] Skip

• Display prompt:

  ⏭️ This file has been skipped, process later

• Immediately proceed to the next file

User selects [Q] Exit

• Display exit prompt:

  👋 Daily review process has ended. [N] files remaining to be processed.

• End daily review process

Decision Criteria (Quick Reference)

Standards for "Keep" (Must meet at least one)

• I can use it now (needed for current projects)
• Need to verify authenticity (others' experiences need practical testing)
• Strongly related to main focus areas:
  • Official account writing (AI, knowledge management, personal growth)
  • Medical device software development (architecture, quality, compliance)
  • Full-stack independent development
• Contains complete problem-solving paths (reproducible)
• Reusable code/configuration
• Contains unique pitfall avoidance guides

Standards for "Delete" (Delete if any is met)

• This knowledge is important (but irrelevant to me)
• Might be useful later (hoarding mentality)
• The author is well-known (but I have no actual need)
• Temporary issues (already solved, no reuse value)
• Duplicate content (better notes already exist)
• Obviously irrelevant content (unrelated to main focus areas)
• Pure information hoarding

Edge Case Handling

• If value cannot be judged within 2 minutes → Default to Delete
• Truly important content will reappear
• Avoid the "just in case" hoarding mentality

references/decision_criteria.md

Execution Specifications (Important)

Interaction Process Principles

1. Scan the inbox to get the list of all pending files
2. Display overview and let user select processing mode:
   • [A] Process all files
   • [P] Partial processing (user selects via serial numbers)
   • [Q] Exit
3. Determine the list of files to process based on user selection
4. For each pending file:
   • Read and analyze
   • Display suggestion and reasons
   • Display options (Y/N/V/S/Q)
   • Stop output, wait for user response
5. Execute corresponding operations after receiving user's reply
6. Automatically proceed to the next file after completing the current one (repeat steps 4-5)
7. Until all selected files are processed or user selects [Q] to exit

• ❌ Display suggestions for all files at once
• ❌ Require batch replies from users (e.g., "YYY" means agree to all)
• ❌ Proceed to next file before user replies
• ❌ Require "Press any key to continue" after processing each file (should proceed directly)
• ❌ Force user to process all files (should allow partial processing)

Agent: [Display overview + processing mode options]
User: P
Agent: [Please enter serial numbers]
User: 1,3,5
Agent: [Confirmation: Will process 3 files + Display suggestion and options for the 1st file]
User: Y
Agent: [Execute operation + Display suggestion and options for the 3rd file]
User: N
Agent: [Ask for further decision]
User: K
Agent: [Execute operation + Display suggestion and options for the 5th file]
...

Batch Processing Principles

• ✅ Support full or partial processing (user's choice)
• ✅ User can flexibly select the number of files to process based on available time
• ✅ Control decision time per file within 30 seconds
• ✅ Complete daily review process in 5-10 minutes (achievable via partial processing)
• ✅ Continuous process, no need to trigger next file manually
• ✅ Unprocessed files can be continued next time, support batch daily review

AI Suggestion Principles

• ✅ Provide clear "Keep/Delete" suggestions (no ambiguity)
• ✅ List 3-5 specific reasons (based on decision criteria)
• ✅ Must provide "suggested move location" when recommending to keep
• ✅ Reasons must be specific, avoid vague statements (e.g., "content is valuable" is too vague)

Decision Traceability Principles

• ✅ All kept files must have decision annotations written
• ✅ Decision reasons are recorded in both frontmatter and top of body content (dual records)
• ✅ Facilitate future audits of "why this was kept at that time"

Manual Execution Principles

• ✅ Users retain final control (manually move/delete files)
• ✅ Avoid the insecurity of "black box automation"
• ✅ Deepen understanding of knowledge flow through manual operations

Collaboration with Other Skills

• conversation-extractor

  : Generate basic notes to the inbox

• knowledge_auditor

  : Conduct in-depth audits on kept files with uncertainty (during weekly reviews)

• Weekly/Monthly Reviews: Conduct in-depth refinement of kept files

• process-doc-generator

  : Upgrade kept practical records to process documents

• qa-appender

  : Append kept Q&A to existing notes

Resource Files

• Decision Criteria: references/decision_criteria.md
• Project Directory Mapping: references/project_directories.md

Differences from Old Version

Old Version (8-option multi-path diversion)

Display: File name + summary + first 200 words preview
↓
User decision: Choose from 8 options (D/K/A/U/M/V/S/Q)
↓
Agent automatically executes: Delete/Move/Append/Upgrade, etc.

• Heavy decision burden (8 options)
• Lack of decision guidance (users need to judge by themselves)
• No decision traceability (disappears after processing)
• Automated black box (worried about misoperations)

New Version (AI suggestions + user confirmation)

Agent analysis: Read complete file + apply decision criteria
↓
Agent suggestion: Provide "Keep/Delete" + 3-5 specific reasons
↓
User confirmation: Agree/Disagree (binary judgment)
↓
If keep: Write decision annotation + modify metadata → User manually moves
If delete: User manually deletes
↓
Proceed to next file

• ✅ Reduced cognitive load (binary judgment vs 8 choices)
• ✅ Improved decision quality (supported by reasons)
• ✅ Enhanced traceability (reasons written in files)
• ✅ Maintained sense of control (manual execution)
• ✅ Decision education (learn AI's judgment criteria)