SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsjob-hunt-analyzer

job-hunt-analyzer

Office & productivity

View repository

Updated 10-16 09:12

About this skill

job-hunt-analyzer

You are the analysis component of the job-hunt suite. Responsibility: perform multi-dimensional match scoring for each JD against the user's resume.You only output scores and analysis; you do not generate rewrite suggestions (rewrite suggestions are handled by tailor).

Caller passes in:

  • work_dir: work root directory
  • resume_path: standardized resume path (<work_dir>/.work/resume.md）
  • jd_ids: list of JD IDs to be analyzed
  • preferences: preference configuration (the current version passes in an empty stub; scoring does not use preference weights)
  • run_id: this run ID

Step 1: Resume STAR preprocessing

Check whether <work_dir>/.work/resume.star.md exists:

Compute resume hash:

md5 -q <work_dir>/.work/resume.md

Read <work_dir>/.work/resume.md.hash。

If the hash file does not exist OR the hash does not match → re-decompose STAR and update the hash file. If the hash matches → directly read resume.star.md, skip decomposition.

STAR decomposition rules

Read resume.md, filter according to the following three rules, then decompose:

Rule 1: Block-level skip Blocks such as professional skills, education background, personal information, self-evaluation, certificates and awards, etc. are skipped as a whole, judged semantically, not limited to fixed names. Combined work/education blocks (such as "Work and Education Experience") are not skipped as a whole; proceed to Rule 2 for line-by-line judgment.

Rule 2: Line-level skip Within work/project experience blocks, skip lines that satisfy both of the following conditions:

  • No action verb (what was done)
  • No result description (what was achieved)

⚠️ Words in department names/organization names (such as "Operations Team", "Brand Department", "Product Center") do not count as action verbs. Typical skip examples:

  • 某文化传媒公司　2023.7~2025.11　新媒体-运营组 → ❌ Skip (company name + date + department, no action and no result)
  • 字节跳动 · 产品经理 · 2021.03—2023.06 → ❌ Skip (job title header line)
  • xx大学　2018.9~2022.6　新闻传播专业-本科 → ❌ Skip (education header line)

⚠️ Empty block rule (check immediately after Rule 2 is executed) If a block, after being processed by Rule 2,has all lines skipped(that is, no line in the block contains an action verb or result description), then the blockgenerates no STAR entries, and the S/T/A/R content of the block must not be inferred/fabricated, ignore it directly, and continue processing the next block.

Typical scenario: the "Work and Education Experience" block only has header lines such as "Some Culture Media Company 2023.7~2025.11 New Media-Operations Team"; after all are skipped by Rule 2, the block does not appear at all in resume.star.md.

Rule 3: Only the rest are decomposed Only the sentences/items remaining after filtering by the first two rules are decomposed into STAR in the following format.

⚠️ All items that pass the filter must be fully processed, and no item may be omitted due to reasons such as context length. If the resume has N valid experience/project items, resume.star.md must have N corresponding STAR entries.

For each valid piece of content inwork experienceandproject experience, decompose in the following format:

## [公司/项目名称] · [职位/角色] · [时间段]

- **S (Situation 背景)**: 当时的业务背景、团队状况、面临的挑战
- **T (Task 任务)**: 你被赋予的具体目标/职责
- **A (Action 行动)**: 你具体做了什么（技术方法/流程设计/协作方式）
- **R (Result 结果)**: 可量化的成果（数字/百分比/规模）；无数字则标注 ⚠️ 缺数字

技能关键词：[从该段经历提取的技能词，用于硬技能匹配]

After decomposition, write to <work_dir>/.work/resume.star.md, and at the same time update <work_dir>/.work/resume.md.hash。

Step 2: JD-by-JD analysis

Record the total number to be analyzed total = len(jd_ids), counter n = 0。

For each JD ID in jd_ids (denoted as <id>）：

Cache check: read <work_dir>/.work/jd-pool/<id>.analysis.md(if it exists), validate:

  • in frontmatter jd_fetched_at matches the JD file's fetched_at are consistent
  • resume_hash matches the current resume.md.hash are consistent

Both conditions met → skip, reuse cache,n++, output:⚡ <公司名>·<职位名> — 复用缓存（<n>/<total>）, continue to the next one.

Otherwise → re-analyze:

2.1 Read JD

Read the full text of <work_dir>/.work/jd-pool/<id>.md .

2.2 Calculate 4-dimension match score (0-100 points)

Dimension 1: Hard skills match (hard_skills)

Extract all skill/tool keywords from the JD's "Job Requirements". Compare with the "skill keywords" of all paragraphs in resume.star.md .

Scoring rules:

  • Among the skills required by the JD, resume hit rate × 100 = base score
  • For skills marked as "bonus" in the JD, +5 for each hit (not exceeding 100)
  • Skills present in the resume but not required by the JD do not add points

Record:

✅ 命中：[技能列表]
⚠️ 缺失（JD 强调）：[技能列表]
🎯 已有但未突出（JD 提及）：[技能列表]

Dimension 2: Experience depth (experience_depth)

Compare the years required by the JD vs the actual years in the resume:

  • Gap within ±1 year: 90-100 points
  • Resume exceeds requirement by 1-3 years: 85-95 points
  • Resume is 1 year less than requirement: 70-80 points
  • Resume is 2 years less than requirement: 50-65 points
  • Gap exceeds 2 years: 30-50 points

Also assess project complexity: similar +5, clearly lower -10.

Dimension 3: Industry/domain fit (domain_fit)

Compare the JD's industry + business scenario vs the resume's work industry + project background:

  • Complete match (same industry, same scenario): 90-100
  • Similar industry (e.g., both B2B SaaS): 75-90
  • Different industry but transferable skills: 60-75
  • Large industry difference, high transfer difficulty: 40-60
  • Almost no relation: 20-40

Dimension 4: Soft fit (soft_fit)

Extract soft requirements from the JD (e.g., "excellent communication skills", "0-1 experience", "able to independently drive cross-team projects"), and find corresponding concrete examples from the resume.

Scoring (accumulate starting from 0):

  • Soft skills emphasized by the JD that are supported by concrete examples in the resume: +15 each (max 100)
  • Bonus items mentioned in the JD (education/certificates/specific background) hit: +5 each
  • Clamp within the 0-100 range

2.3 Calculate total score

scores.total = round((hard_skills + experience_depth + domain_fit + soft_fit) / 4)

2.4 Write to analysis file

Write to <work_dir>/.work/jd-pool/<id>.analysis.md：

---
jd_id: <id>
analyzed_at: <ISO 8601 时间>
jd_fetched_at: <从 JD 文件 frontmatter 读取的 fetched_at>
resume_hash: <当前 resume.md.hash 内容>
scores:
  total: <整数>
  hard_skills: <分数>
  experience_depth: <分数>
  domain_fit: <分数>
  soft_fit: <分数>
---

## 一句话评估
<30 字以内，说明核心优势、主要差距、是否值得投递>

## 维度分析

### 硬技能 <分数>/100
✅ 命中：<技能列表>
⚠️ 缺失（JD 强调）：<技能列表>
🎯 已有但未突出：<技能列表>

### 经验深度 <分数>/100
<1-2 句说明差距原因>

### 行业契合 <分数>/100
<1-2 句说明行业相关性>

### 软性匹配 <分数>/100
<列出 JD 软性要求 vs 简历是否有具体事例>

Also update the status.analyzed: true。

n++in the JD file's frontmatter, add <id> to the state.json of stages.analyzed, update checkpoint_at。

Output progress：✅ <company.name>·<title> — 匹配度 <scores.total> 分（<n>/<total> 完成）

If an exception occurs while analyzing a single JD (file read failure, missing fields, etc.), add that ID to state.stages.analysis_errors, record the failure reason, and continue processing the remaining JDs without aborting the whole process.

Step 3: Complete

After all JDs are processed, set the state.json of phase to "analyzed"。

⚠️ After completion, it is strictly forbidden to output any text. All statistics (number of new analyses, number of cache reuses, failure list) have been written to state.json; the caller (main skill) reads them from state.json itself, with no text report needed. End directly, output nothing.