AgentSkillsCN

batch-processing

利用 Gemini 的 200 万 token 上下文窗口,处理超大型文档或代码库。适用于处理超出常规上下文限制的内容,或需要一次性全面审视所有信息时使用。

SKILL.md
--- frontmatter
name: batch-processing
description: Process multiple files through Gemini workflows for classification, extraction, or analysis. Use when user needs the same operation applied to many files.
allowed-tools: gemini-api, Read, Glob, Write

Batch Processing with Gemini

When to Use This Skill

Automatically invoke this skill when:

  • User needs to process multiple files with the same operation
  • Task involves classification of many items
  • Extracting specific information from multiple documents
  • Analyzing a collection of similar files
  • User requests batch operations on images, CSVs, PDFs, or code files

Examples That Trigger This Skill

  • "Classify all images in this folder"
  • "Extract key points from all meeting notes"
  • "Summarize each customer feedback file"
  • "Find potential issues in all code files"
  • "Analyze sentiment in all review documents"
  • "Generate descriptions for all product images"

How to Use

  1. Parse request: Identify the file pattern and operation
  2. Find files: Use Glob to locate all matching files
  3. Process each file:
    • Read file content
    • Determine appropriate Gemini tool (analyze_visual for images, process_long_context or gemini_query for text)
    • Call Gemini with the operation prompt
    • Collect result
  4. Aggregate results: Combine outputs with file-by-file breakdown
  5. Create summary: Identify patterns or insights across all results
  6. Optional: Save results to file if user requests

Tool Selection

  • Images/Videos: Use analyze_visual tool
  • Text files < 50K tokens each: Use gemini_query tool
  • Text files > 50K tokens: Use process_long_context tool
  • Mixed content: Adapt per file type

Output Format

Structure results clearly:

markdown
# Batch Processing Results

**Files Processed**: X
**Operation**: [description]
**Timestamp**: YYYY-MM-DD HH:MM

## Results by File

### filename1.ext
[Gemini's response for this file]

### filename2.ext
[Gemini's response for this file]

## Aggregate Analysis

[Overall patterns, common themes, summary insights]

Capabilities

  • Parallel Processing: Can process multiple files efficiently
  • Mixed Content Types: Handle images, text, code, data files
  • Structured Output: Generate organized results for easy review
  • Pattern Detection: Identify trends across batch
  • Export Options: Save results to CSV, JSON, or Markdown

Best Practices

  • Show progress to user ("Processing file 5 of 20...")
  • Use gemini-1.5-flash for speed when processing many files
  • Batch similar file types together when possible
  • Create clear result structure for easy navigation
  • Offer to save results to file for large batches
  • Provide aggregate summary in addition to individual results

Error Handling

  • Continue processing remaining files if one fails
  • Report errors clearly with file name
  • Provide partial results even if some files fail
  • Suggest fixes for common errors (unsupported format, file not found)

Example Workflow

code
User: "Classify all images in screenshots/ folder"

1. Glob: screenshots/*.{jpg,png}
2. Found: 15 images
3. For each image:
   - analyze_visual(file_path, "Classify this image: UI mockup, diagram, screenshot, or other")
4. Collect all classifications
5. Summary: "10 UI mockups, 3 diagrams, 2 screenshots"
6. Present structured results