process-file（过程文件）

Process File Skill

Generic file processing Skill supporting multiple file formats for parsing and intelligent archiving, fully integrated with the AkashicRecords governance system.

When to use this Skill

• User says "read", "process"
• User says "archive", "import"
• User provides file path for processing
• User wants to integrate external files into knowledge base
• User provides email, PDF, Office documents, images, etc.

Workflow

1. Initialization - Read Preferences

Check claude.md:

1. Read current project's claude.md
2. Look for file-handling-preferences related record
3. If path found, read preferences file

If no preferences file exists:

1. Ask user: "This is the first time using process-file skill in this project. Where would you like to create the file handling preferences?"
2. Suggest default location: file-handling-preferences.md in project root
3. After user confirmation, create file and record location in claude.md

Preferences file structure:

• Processing pattern records (by file type and content category)
• Auto processing settings (whether to allow saving without confirmation)
• Historical processing records

2. File Type Detection

Detect file type: Determine processing method based on file extension:

Tool availability check:

• Check if required tools are installed before execution
• If mu not installed: Prompt Please install maildir-utils: sudo apt install maildir-utils
• If markitdown not installed: Prompt Please install markitdown: pip install markitdown

3. Content Extraction

Email (.eml):

mu view <filepath>

Extract: sender, recipient, subject, date, body

PDF/Office documents:

markitdown <filepath>

Convert to markdown format

Images: Use Read tool to directly read image, let language model analyze content:

• Identify image subject
• Extract text (if any)
• Describe image content

Audio/Video:

1. Ask user for suggested processing method
2. Possible options:

- Record file metadata only - Use external tool for transcription - Record manual summary

1. Record user's chosen processing method in project claude.md

4. Content Analysis

Analyze content:

• Identify topics and keywords
• Determine content type (technical, personal, work, academic, etc.)
• Extract important information (dates, people, places, events)

Infer user intent:

• Archive for storage (long-term preservation)
• Project update (related to existing project)
• Record memo (personal notes)
• Data organization (batch processing)

Match against preferences:

• Check if preferences file has matching patterns
• If historical records exist, prioritize suggesting same processing method

5. Directory Discovery

Use akashicrecords mechanism:

1. Based on content analysis results, build search query
2. Scan knowledge base directory structure
3. Read each directory's RULE.md to understand purpose
4. Evaluate content-to-directory purpose match

Suggestion logic:

• Technical document + directory purpose "research" → high match
• Email + directory purpose "communications" → high match
• Personal photo + directory purpose "personal life" → high match
• No clear match → suggest Miscellaneous or ask user

6. User Confirmation

Present analysis results:

## File Analysis Results

**File**: [filename]
**Type**: [file type]
**Content Summary**: [brief summary]

**Inferred Intent**: [archive/update/record]

**Suggested Location**: [target directory path]
**Reason**: [why this location was chosen]

**Planned Operation**:
- Call add-content skill
- Format: [according to RULE.md]
- Filename: [suggested filename]

Do you approve this operation?

Wait for confirmation:

• Default requires user approval
• If auto_save: true in preferences, can skip confirmation
• User can modify suggested location or cancel

7. Execute

Call corresponding akashicrecords skill:

• Add new content → add-content skill
• Update existing → update-content skill

Format according to target RULE.md:

• Read target directory's RULE.md
• Follow naming conventions
• Apply frontmatter format (if required)

8. Update Preferences

Record this processing experience:

### [Date] [File Type]
- Content characteristics: [key features]
- Target location: [actual storage location]
- Processing method: [skill used]

Learning pattern:

• Accumulate user preferences
• Prioritize suggesting same method for similar content next time

Multi-File Processing

When user provides multiple files:

Parallel Analysis

1. Launch a subagent for each file
2. Each subagent independently executes Phase 2-5
3. Wait for all subagents to complete

Consolidated Presentation

## Multi-File Processing Analysis Results

| # | Filename | Type | Content Summary | Suggested Location | Operation |
|---|----------|------|-----------------|-------------------|-----------|
| 1 | file1.pdf | PDF | [summary] | Research/ | add-content |
| 2 | photo.jpg | Image | [summary] | Personal/ | add-content |
| 3 | email.eml | Email | [summary] | Work/ | add-content |

Please choose:
- Approve all
- Confirm individually
- Cancel

Batch Execution

• After user approves all, execute sequentially
• When user confirms individually, confirm each file separately

Error Handling

Tool Not Installed

Warning: Cannot process .eml file: mu tool not installed
Please run: sudo apt install maildir-utils

Unsupported File Format

Warning: Unsupported file format: .xyz
How would you like to proceed?
1. Try reading as plain text
2. Record file metadata only
3. Skip this file

Parse Failure

Warning: Unable to parse file content
Error: [error message]
How would you like to proceed?
1. Retry
2. Enter summary manually
3. Skip this file

Integration with Governance

Before operation:

• Read preferences file
• Confirm akashicrecords governance structure exists

During operation:

• Use akashicrecords skills for actual operations
• Follow target directory's RULE.md

After operation:

• Update preferences file
• akashicrecords skills automatically handle README.md updates

Examples

Example 1: Process PDF Paper

User: "Read ~/Downloads/transformer-paper.pdf"

Workflow:

1. Check preferences → Find historical record "technical paper → Research/Papers/"
2. Detect.pdf → Use markitdown
3. Execute markitdown ~/Downloads/transformer-paper.pdf
4. Analyze content → AI/machine learning topic
5. Match preferences → Matches "technical paper" pattern
6. Suggest Research/Papers/AI/
7. User confirms
8. Call add-content skill
9. Update preferences file

Example 2: Batch Process Emails

User: "Archive these emails: email1.eml email2.eml email3.eml"

Workflow:

1. Detect multiple files → Launch 3 subagents
2. Each subagent processes in parallel:

- Parse using mu view - Analyze sender, subject, content - Suggest target location

1. Consolidate results into list
2. User selects "Approve all"
3. Execute add-content sequentially
4. Update preferences

Example 3: Process Image

User: "Process this photo ~/Photos/vacation.jpg"

Workflow:

1. Detect.jpg → Use Read tool
2. Language model analyzes image content → "Beach vacation photo"
3. Check preferences → Find "travel photos → Personal/Travel/"
4. Suggest Personal/Travel/2025/
5. User confirms
6. Call add-content (convert to descriptive markdown)
7. Update preferences

Best Practices

1. Always check preferences first - Prioritize historical processing patterns
2. Confirm before saving - Default requires user approval
3. Update preferences after success - Accumulate learning user preferences
4. Use parallel processing - Leverage subagents for multiple files
5. Handle errors gracefully - Provide alternatives
6. Integrate with akashicrecords - Use existing skills for operations

Notes

• Preferences file path is recorded in project claude.md
• Each project can have different preferences
• Audio/video processing methods are recorded in claude.md
• This Skill does not modify files directly, operates through akashicrecords skills