AgentSkillsCN

linkedin-job-filter

LinkedIn 职位自动筛选工具。借助带有头部参数的 Agent-Browser 打开 LinkedIn 职位搜索页面，逐个点击职位查看完整职位描述，调用职位筛选工具进行评估，并对未通过筛选的职位点击“关闭”（X）按钮。触发方式包括：“筛选 LinkedIn 职位”、“LinkedIn 职位筛选”，或当用户提供了 LinkedIn 职位搜索的 URL 时。

SKILL.md
--- frontmatter
name: linkedin-job-filter
description: LinkedIn job auto-filter tool. Uses agent-browser (--headed) to open LinkedIn job search pages, clicks each job to view full JD, calls jd-filter for evaluation, and clicks dismiss (X) button for rejected jobs. Triggers include "filter LinkedIn jobs", "linkedin job filter", or when user provides a LinkedIn Jobs search URL.
allowed-tools: Bash(agent-browser:*), Skill(jd-filter)

LinkedIn Job Filter Skill

Automatically filter LinkedIn job search results by viewing each job detail, using jd-filter for evaluation, and auto-dismissing jobs that don't meet criteria.

Use Cases

  • •User provides LinkedIn Jobs search results URL (e.g., https://www.linkedin.com/jobs/search/...)
  • •User says "help me filter LinkedIn jobs" or "filter LinkedIn jobs"
  • •User has already searched jobs on LinkedIn and wants to batch filter

Prerequisites

  • •agent-browser installed and in PATH
  • •User logged into LinkedIn (requires agent-browser state or manual login)
  • •jd-filter skill's user_filter.json configured

Limitations

  • •No pagination support - Only processes jobs visible on current page
  • •Requires user to be logged into LinkedIn (if not logged in, prompt user to login manually)

Workflow

Step 0: Load Filter Criteria

Read jd-filter's filter criteria:

code
.claude/skills/jd-filter/user_filter.json

If doesn't exist, prompt user to set filter first (using jd-filter skill).

Step 1: Open LinkedIn Page

bash
agent-browser --session linkedin --headed open "<linkedin-jobs-url>"
agent-browser --session linkedin wait --load networkidle
agent-browser --session linkedin wait 3000

Important: Must use --headed, user needs to see browser window.

Step 2: Check Login Status

bash
agent-browser --session linkedin snapshot -i

Check snapshot output:

  • •If see "Sign in" / "Join now" buttons → Prompt user to login manually
  • •If see job listings → Continue to next step

If login required:

  1. •Prompt user to manually complete login in browser
  2. •Wait for user to confirm logged in
  3. •Re-navigate to job search URL
  4. •Snapshot again to confirm

Step 3: Get Job Listings

bash
agent-browser --session linkedin snapshot -i

From snapshot, identify all job cards. LinkedIn search results typically contain:

  • •Job title link (clickable)
  • •Company name
  • •Location
  • •Posted time
  • •Dismiss button (X icon)

Record for each job:

  • •title_ref: Job title's @ref (for clicking to view details)
  • •dismiss_ref: X button's @ref (for dismissing unqualified jobs)
  • •title_text: Job title text
  • •company_text: Company name

Save all job info to in-memory list, ready to process one by one.

Step 4: Process Each Job

For each job in the list, execute the following workflow:

4a: Click Job to View Details

bash
agent-browser --session linkedin click @title_ref
agent-browser --session linkedin wait --load networkidle
agent-browser --session linkedin wait 2000

After clicking, LinkedIn right panel will display job details.

4b: Extract JD Text

bash
agent-browser --session linkedin snapshot -i

From snapshot, find job description area. LinkedIn JD is usually in:

  • •Under jobs-description__content class
  • •Or under jobs-box__html-content class

After finding JD content's @ref:

bash
agent-browser --session linkedin get text @jd_ref

Also extract:

  • •Job title
  • •Company name
  • •Location
  • •Job type (Full-time, Contract, etc.)
  • •Remote/Hybrid/On-site

4c: Call jd-filter for Evaluation

Pass extracted JD text to jd-filter for evaluation. jd-filter will return:

  • •PASS - Job meets user requirements
  • •REJECT - Job doesn't meet user requirements

4d: Handle Filter Results

If REJECT:

  1. •Find the dismiss (X) button for this job in current snapshot
  2. •Need to re-snapshot to get latest refs:
    bash
    agent-browser --session linkedin snapshot -i
  3. •Find dismiss button's @ref (usually a small X icon button in top-right corner of job card)
  4. •Click dismiss:
    bash
    agent-browser --session linkedin click @dismiss_ref
agent-browser --session linkedin wait 1000
  5. •Output: REJECT - [Job Title] @ [Company] - [Rejection Reason]

If PASS:

  1. •Take no action, keep the job
  2. •Output: PASS - [Job Title] @ [Company] - [Match Reason]

4e: Prepare for Next Job

bash
agent-browser --session linkedin snapshot -i

Re-get snapshot, because dismiss action may have changed page structure and refs. Find next unprocessed job, repeat 4a-4d.

Step 5: Output Summary Report

After all jobs processed, output summary:

code
## LinkedIn Job Filter Report

Total Processed: X jobs
- PASS: Y jobs
- REJECT: Z jobs

### PASS Jobs:
1. [Job Title] @ [Company] - [Location]
2. ...

### REJECT Jobs:
1. [Job Title] @ [Company] - [Rejection Reason]
2. ...

Step 6: Keep Browser Open

Don't close browser - Keep browser open after processing, user may need to continue operations.

LinkedIn Page Structure Reference

Job Listings Page

Typical LinkedIn search results page structure:

  • •Left side: Job card list (scrollable)
  • •Right side: Selected job's detailed information panel

Common Elements

ElementDescriptionHow to Find
Job TitleClickable link<a> tag in snapshot, text is job name
Company NameTextBelow job title
LocationTextBelow company name
Dismiss ButtonX iconButton in top-right corner of job card, usually has dismiss related attribute
JD ContentDetail panelMain text area in right panel
"Show more" ButtonExpand full JDBottom of JD area

Expand Full JD

LinkedIn shows only partial JD by default, need to click "Show more" / "See more" to expand:

bash
# Find "Show more" button in JD panel
agent-browser --session linkedin snapshot -i
# If found "Show more" / "...more" / "See more" button
agent-browser --session linkedin click @show_more_ref
agent-browser --session linkedin wait 1000
agent-browser --session linkedin snapshot -i
# Now can get full JD
agent-browser --session linkedin get text @full_jd_ref

Key Principles

  1. •Must use --headed - User needs to see browser
  2. •Re-snapshot after each action - LinkedIn is SPA, DOM changes frequently
  3. •Moderate speed - Wait 1-2 seconds between jobs, avoid rate limiting
  4. •Don't stop on errors - If job extraction fails, log error and continue to next
  5. •Snapshot before action - Never use stale @ref
  6. •Expand JD - Click "Show more" to get full description before filtering
  7. •Re-locate after dismiss - After dismissing a job, list changes, need to re-snapshot
  8. •Don't close browser - Keep browser open after processing, user may continue

Processing Efficiency Optimization

Efficient Processing Strategy

  • •Obviously irrelevant jobs: For retail, part-time, healthcare etc. that are clearly unrelated, can dismiss based on title alone without reading full JD
  • •Prioritization: Only jobs that look promising from title need full JD reading, saves time

Behavior After Dismiss

  • •After dismissing a job, LinkedIn auto-shows new job to fill the gap
  • •Must track newly appeared jobs - Don't miss jobs that appear after dismiss
  • •When re-snapshotting, check list and process all unprocessed jobs

Error Handling

Error TypeHandling
Not logged inPrompt user to login manually, wait for confirmation
Job details load failureWait 3 seconds and retry once, still fails then skip
JD text emptyTry clicking "Show more", still empty then skip
Dismiss button not foundSkip dismiss, mark for manual processing
Page rate limit/CAPTCHAPause, prompt user to complete verification then continue
Snapshot timeoutWait 5 seconds then retry

Usage Example

code
User: Help me filter jobs on this LinkedIn page https://www.linkedin.com/jobs/search/?keywords=software%20engineer&location=Toronto