zhuangwenhui/autumnrecuitmentskills/recruiting-jd-sync

Recruiting JD Sync

Use this skill to run a durable recruiting workflow:

1. collect job list entries and authoritative detail pages,
2. deduplicate against the current inventory,
3. store complete JD text in a dictionary file,
4. classify jobs by technical-stack depth and resume availability,
5. synchronize the application workbook,
6. validate counts, IDs, links, styles, and grouping rules.

Core Contract

Maintain three artifact roles unless the user specifies another architecture:

• JD dictionary Markdown: complete original JD text, URL, source, city, salary, dates, and notes.
• Classification Markdown: technical categories, inventory IDs, resume-version mapping, and application judgment.
• Application workbook: practical tracker for application actions, source, URL/contact, recommended resume, and company/group sheets.

Use 入库编号 as the only formal cross-file key. Do not add or preserve a second 原编号 field unless the user explicitly requests a historical migration table.

Workflow

1. Preflight

• Inspect the live file tree before assuming filenames or folder structure.
• If browser login state is needed, read references/browser-preflight.md before controlling the browser.
• If an .xlsx workbook is involved, load workbook dependencies and plan a verification pass.

1. Collect

• Treat list pages as indexes only. Use job detail pages as the authoritative JD source.
• Extract URL, company, job title, city, salary, education, source, refresh/date information, and full JD body.
• Do not refresh or reset an already opened user page unless the user explicitly asks.

1. Deduplicate

• Read references/dedupe-policy.md.
• Prefer strong URL/job-id dedupe, then company-title-city, then JD-body comparison.
• For same-title roles with different links, compare detail text before deciding whether to keep both.

1. Classify And Number

• Read references/classification-policy.md.
• Classify from the actual JD technical requirements, not from fixed category names.
• Use D{category}-{sequence} IDs. Category count is determined by JD clusters and available resume variants.
• If resume count is lower than category count, preserve the categories and use resume placeholders unless the user asks to merge categories.

1. Update Files

• Update the JD dictionary first.
• Update the classification file second.
• Update the workbook last, using grouped sheets for special companies when the workbook already has that convention.
• In the workbook, only web URLs should be hyperlinks. Emails and WeChat IDs must remain plain text.

1. Validate

• Run the scripts in scripts/ where applicable.
• Required checks: equal ID sets across files, no duplicate IDs, no stale 原编号, no orphan links, no bad web-link targets, no unintended submitted fill on new rows, workbook style consistency.

References

• Browser and permissions: references/browser-preflight.md
• Classification and resume mapping: references/classification-policy.md
• Workbook rules: references/workbook-rules.md
• Deduplication: references/dedupe-policy.md

Scripts

• scripts/validate_artifacts.py: compare IDs and basic invariants across Markdown and workbook files.
• scripts/scan_workbook.py: report sheet counts, sources, styles, hyperlinks, group hits, and submitted-fill issues.
• scripts/normalize_xlsx_style.py: normalize workbook row style, filters, freeze panes, and optional orphan blank rows.
• scripts/rebuild_xlsx_links.py: rebuild workbook link-column hyperlinks from the JD dictionary by inventory ID.
• scripts/extract_resume_variants.py: inspect available resume PDFs and infer resume/category mapping state.

Prefer running scripts from the project/workspace directory and pass explicit paths when filenames differ.