recruitment-processor

Recruitment Information Processing Skill

Task Objective

This Skill is used to: batch process markdown documents containing recruitment information, automatically identify image content (including local images and online image links), extract key information, filter according to conditions, and generate a structured report
Capabilities include: file path identification and scanning, adaptive batch processing, image link identification, image download, image recognition and OCR, text understanding, information extraction, condition matching, progress tracking, completeness verification, report generation
Trigger conditions: the user provides recruitment markdown documents (a single file or a folder) and filtering conditions

Prerequisites

Dependency description: the image download function depends on the requests library, version >= 2.28.0

Operation Steps

Standard Process

Phase One: File Scanning and Planning

Input Path Identification
- Determine whether the path provided by the user is:
  - Single file: add it directly to the processing list
  - Folder: scan all markdown files in the folder
- Record the file type and path
File Scanning (for folders)
- If it is a folder, scan all markdown files in that directory
- Supported file extensions:.md, .markdown
- Recursively scan subdirectories (optional, determined according to user needs)
- Record the full paths of all files
- All files must be scanned, none may be omitted
Quantity Assessment and Batching Decision
- Count the total number of documents that need to be processed
- Adaptive Batching Rules：
  - Number of documents <= 10: process all documents at once
  - Number of documents > 10: process in batches, with 10 documents per batch
- Calculate how many batches are needed
- Record the batch plan: total batches, file list for each batch
Progress Tracking Initialization
- Record the following information:
  - Total number of files
  - Number processed (initially 0)
  - Number pending (initially the total number of files)
  - Current batch (initially 0)
  - Number successfully processed (initially 0)
  - Number failed to process (initially 0)

Phase Two: Batch Process Documents

Batch Loop Processing For each batch (executed in order):

5.1 Batch Initialization
- Current batch number += 1
- Determine the file list to be processed in this batch (take 10 files in numerical order or the remaining files)
- Report the progress of this batch: "Processing batch X, out of Y batches, with Z files"
5.2 Process Documents One by One For each recruitment document in this batch,it must be processed completely, and must not be skipped：

5.2.1 Image Link Identification and Extraction
- Scan all image references in the document and identify two formats:
  - Local images:![alt](local_path.png)
  - Online images:![alt](https://example.com/image.jpg)
- Extract the URL or path of all images
- Record the position and context of images in the document
5.2.2 Online Image Download For each online image link:
- Call python /workspace/projects/recruitment-processor/scripts/download_image.py --image-url <URL> --output-dir ./downloaded_images
- Capture the download result and obtain the local save path
- If the download fails, record the error but continue processing other images
- Record the image source (online download/local file)
5.2.3 Image Recognition and Content Extraction
- For all accessible images (successfully downloaded online images and local images)
- Use image recognition capabilities to extract the text information from each image
- All images must be recognized one by one, without omitting any
- Record the correspondence between image content and context
- Label the information source of each image
5.2.4 Text Understanding and Integration
- Read the document text content
- Integrate and understand the text information recognized from images with the text content
- Ensure that both text and image information are fully utilized
- Form a complete understanding of the recruitment document
5.2.5 Key Information Extraction Extract the following information (if a certain piece of information is not explicitly mentioned in the document, mark it as "not mentioned"):
- Job title
- Company name
- Salary and benefits (explicit range or specific value)
- Deadline
- Publication time
- Enterprise nature (state-owned enterprise/private enterprise/foreign enterprise/public institution, etc.)
- Work location
- Educational requirements
- Work experience requirements
- Job responsibilities
- Job requirements
- Statistics on the number and sources of images
5.2.6 Error Handling
- If a file fails to be processed:
  - Record the reason for failure
  - Failed processing count += 1
  - Do not affect the processing of other files; continue processing the remaining files in the batch
- If the file is processed successfully:
  - Successful processing count += 1
  - Temporarily store the extracted information
5.3 Batch Progress Update
- Processed count += number of files in this batch
- Pending count = total number of files - processed count
- Report batch progress: "Batch X completed, A files processed, B files pending"

Stage 3: Conditional Filtering and Summary

Conditional Filtering
- Read the filtering conditions provided by the user (refer to the filtering condition template format)
- Perform condition matching on all successfully extracted recruitment information
- Strictly determine whether all filtering conditions are met
- Keep all information that meets the conditions and remove those that do not
- Every document must undergo filtering judgment and must not be skipped
Integrity Verification
- Verify the processing results:
  - Processed count + pending count = total number of files
  - Every file has a processing result (success or failure)
- If omissions are found:
  - Immediately supplement and process the omitted files
  - Update progress tracking information
- Must ensure all files are reviewed, with no omissions allowed
Generate summary document Generate a structured summary report for all filtered recruitment information, including:

8.1 Report Overview
- Total number of documents processed
- Batch processing status: total number of batches, number of files per batch
- Number of successfully processed documents
- Number of failed documents and list of failure reasons
- Total number of extracted images (online images/local images)
- Number of items meeting the filtering criteria
- Processing time
8.2 Recruitment Information List For each recruitment information item that meets the criteria, present it in the following format:
```
## [职位名称] - [公司名称]

**基本信息**
- 薪资待遇：[具体数值或范围]
- 企业性质：[国企/民企/外企等]
- 工作地点：[城市/地区]

**时间信息**
- 截止时间：[具体日期]
- 发布时间：[具体日期]

**要求信息**
- 学历要求：[学历层次]
- 工作经验：[年限要求]

**岗位详情**
- 岗位职责：[摘要]
- 任职要求：[摘要]

**图片信息**
- 图片总数：[数量]张（在线图片X张，本地图片Y张）
- 图片来源标注：[哪些关键信息来源于图片]

---
```
8.3 Failed File List
- List all files that failed processing
- Explain the failure reason for each file
- Provide possible solutions
8.4 Processing Progress Report
- Batch processing timeline
- List of files processed in each batch
- Progress tracking information
8.5 Notes
- Indicate whether information comes from images or text
- Mark missing key information
- Explain the status of image download and recognition
Output Results
- Output the summary document in markdown format
- File name format: Recruitment Information Summary_YYYYMMDD_HHMMSS.md

Resource Index

Required script: see scripts/download_image.py(Purpose: download online images from URL, parameters: --image-url --output-dir <directory>)
Filtering criteria template: see references/筛选条件模板.md(Purpose: define the standard format for filtering criteria)
Information extraction guide: see references/信息提取指南.md(Purpose: guide how to accurately extract key information)

Notes

Completeness requirements (highest priority)

Must completely read all provided documents; skipping any document is strictly prohibited
Must identify all images in the documents one by one, without missing any
Must process all image links, including local images and online images
Must ensure all files are reviewed, with no omissions allowed
After processing is completed, completeness verification must be performed

Adaptive batch processing

Automatically select the processing strategy based on the number of documents:
- Few documents (≤10): process at once
- Many documents (>10): process in batches, 10 per batch
During batch processing, execute strictly in order and do not skip any batch
Report progress after each batch is processed
If a batch fails, it does not affect the processing of subsequent batches
After all batches are completed, perform filtering and aggregation uniformly

Progress tracking

Track processing progress in real time
Record the processing status of each file
Update progress information after each batch is completed
Facilitate monitoring of overall processing status and troubleshooting

Error handling

Failure to process a single file does not affect other files
Record all failed files and the reasons for failure
List failed files separately in the summary document
Provide the error reasons and possible solutions

Image link processing

Identify image syntax in markdown:
- Distinguish between local image paths and online URLs
- Online images must be downloaded locally before recognition
- Images that fail to download should have the error recorded but should not affect overall processing

Image recognition accuracy

Carefully verify the consistency between the text extracted from the image and the context, and perform recognition multiple times if necessary to ensure accuracy

Strictness of conditional filtering

Every condition proposed by the user must be strictly satisfied; if any part does not meet it, it is considered non-compliant

Handling of missing information

For information not explicitly mentioned in the document, it must be marked as "not mentioned"; do not make subjective inferences

Unified time format

All time information should be unified in the YYYY-MM-DD format

Salary format specification

Salary information should include the numerical range and unit (e.g., 15-25K/month, annual salary 300,000-400,000)

Usage examples

Example 1: Processing a single file

Function description: The user provides a single markdown file
Execution method: One-time processing
Key points：
- Directly read the file content
- Fully process all images
- Extract key information
- Perform conditional filtering

Filtering conditions：

薪资范围：15-25K/月
企业性质：国企

Example 2: Processing a small number of files (≤10)

Function description: The user provides a folder containing 5 markdown files
Execution method: One-time processing
Key points：
- Scan all markdown files in the folder
- Number of documents ≤10, process in one go
- Process each file one by one
- Perform filtering and summarization uniformly

Filtering conditions：

工作地点：北京
学历要求：本科及以上

Example 3: Processing a large number of files (>10) - batch processing

Function description: The user provides a folder containing 25 markdown files
Execution method: Batch processing (3 batches: 10+10+5)
Key points：
- Scan the folder and identify 25 files
- Number of documents >10, automatically batch
- Batch 1: Process files 1-10
- Batch 2: Process files 11-20
- Batch 3: Process files 21-25
- Report progress after each batch is completed
- Perform unified filtering after all batches are completed
- Ensure all files have been reviewed

Filtering conditions：

薪资范围：15-25K/月
企业性质：国企

Example 4: Processing a document containing online images

Function description: The document contains online image links, which need to be downloaded before recognition
Execution method: The agent calls a script to download images + the agent performs recognition
Key points：
- Identify image URLs in markdown
- Call download_image.py to download images
- Recognize the content of the downloaded images
- Record that the image source is "online download"

Filter conditions：

薪资范围：15-25K/月
企业性质：国企

Example 5: Multi-condition combined filtering

Function description: Filter technical positions in the Beijing area with a bachelor's degree and 1-3 years of experience
Execution method: Agent processing
Key points：
- Location information may be in the body text or in images
- Work experience requirements may be expressed vaguely (such as "new graduates may apply" or "those with relevant experience preferred"), and need to be judged accurately
- Education requirements may have multiple levels; select the lowest requirement as the filtering basis
- Identify all images one by one to ensure no key information is missed
- Strictly process all files according to the batching logic

Filter conditions：

工作地点：北京
学历要求：本科及以上
工作经验：1-3年
职位类型：技术岗位

Example 6: Processing documents with mixed image sources

Function description: The document contains both local images and online image links
Execution method: The agent calls a script to download online images + the agent recognizes all images
Key points：
- Distinguish between local images and online images
- Download all online images
- Identify all images one by one (local + downloaded)
- Indicate the image source in the summary
- Ensure all images are recognized
Filter conditions：
```
无特殊条件，仅提取信息
```

Example of batch processing workflow

Scenario: Processing a folder of 30 files

Phase 1: File scanning and planning

Scan the folder and find 30 markdown files
Number of documents > 10, decide to process in batches
Calculate batches: 30 files, 10 per batch, 3 batches in total
Initialize progress: total files = 30, processed = 0, pending = 30

Phase 2: Batch processing

Batch 1 processing

Current batch = 1, process files 1-10
Scan files 1-10, recognize images, extract information
Processing result: 10 succeeded, 0 failed
Update progress: processed = 10, pending = 20
Report: "Batch 1 completed, 10 files processed, 20 files pending"

Batch 2 processing

Current batch = 2, process files 11-20
Scan files 11-20, recognize images, extract information
Processing result: 9 succeeded, 1 failed (file 15 corrupted)
Update progress: processed = 20, pending = 10
Report: "Batch 2 completed, 20 files processed, 10 files pending"

Batch 3 processing

Current batch = 3, process files 21-30
Scan files 21-30, recognize images, extract information
Processing result: 10 succeeded, 0 failed
Update progress: processed = 30, pending = 0
Report: "Batch 3 completed, 30 files processed, 0 files pending"

Phase 3: Conditional filtering and summary

Perform conditional filtering on the 29 successfully processed recruitment information items
Completeness verification: processed 30 = total files 30, verification passed
Generate a summary document, including:
- Processing overview: total 30, 29 succeeded, 1 failed
- List of recruitment information meeting the conditions
- Failed file list (File 15: file corrupted)
- Batch processing progress report

Image download script usage instructions

Invocation method

python /workspace/projects/recruitment-processor/scripts/download_image.py --image-url <图片URL> --output-dir <输出目录>

Parameter description

--image-url: Required, the URL address of the image
--output-dir: Optional, the image save directory, defaults to./downloaded_images

Return result

Success: outputSUCCESS:<本地路径>
Failure: outputERROR:<错误信息>to the standard error stream

Use cases

When the document contains online image links (such ashttps://example.com/image.jpg), first call this script to download the image, then perform image recognition on the downloaded local file.

Example

# 下载图片到默认目录
python /workspace/projects/recruitment-processor/scripts/download_image.py --image-url https://example.com/salary.jpg

# 下载图片到指定目录
python /workspace/projects/recruitment-processor/scripts/download_image.py --image-url https://example.com/salary.jpg --output-dir ./temp_images

About this skill

Task Objective

Prerequisites

Operation Steps

Standard Process

Phase One: File Scanning and Planning

Phase Two: Batch Process Documents

Stage 3: Conditional Filtering and Summary

Resource Index

Notes

Completeness requirements (highest priority)

Adaptive batch processing

Progress tracking

Error handling

Image link processing

Image recognition accuracy

Strictness of conditional filtering

Handling of missing information

Unified time format

Salary format specification

Usage examples

Example 1: Processing a single file

Example 2: Processing a small number of files (≤10)

Example 3: Processing a large number of files (>10) - batch processing

Example 4: Processing a document containing online images

Example 5: Multi-condition combined filtering

Example 6: Processing documents with mixed image sources

Example of batch processing workflow

Scenario: Processing a folder of 30 files

Image download script usage instructions

Invocation method

Parameter description

Return result

Use cases

Example