Skip to content
guiyangyuanPublic

About

Local-first Codex skill for Mainland China job discovery and application tracking.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Latest commit

 

History

3 Commits

Folders and files

Repository files navigation

Job Radar

A local-first Codex skill that discovers Mainland China job openings,
explains why they match, and keeps every application status under your control.

Tests Python 3.11+ Local first MIT License

Job Radar recommendation dashboard

Real Job Radar UI with fully synthetic companies, roles, and application data.

Why Job Radar?

Job hunting usually splits into two disconnected problems: finding roles that are actually open, and remembering what happened after clicking Apply. Job Radar connects both without uploading your resume or application history to another service.

  • Find current roles from employer career pages and other public Mainland China recruitment sources.
  • Keep useful links by distinguishing an exact job detail page, an application flow, a searchable listing, and a generic careers homepage.
  • Choose one primary role per company with at most two alternatives, instead of flooding the dashboard with near-duplicates.
  • Respect application limits when an official company source publishes a quota or role-change rule.
  • Track the real funnel from applied and assessment through interview, offer, rejection, withdrawal, and archive.
  • Review recruitment emails locally through read-only IMAP, then confirm each proposed progress change before it becomes canonical.
  • Keep personal data local in explicit JSON files that remain outside the installed skill and outside Git.

Dashboard

Company-first recommendations

Each company gets one primary recommendation, up to two alternatives, a direct destination label, and an explicit application-limit badge. Unknown rules stay unknown instead of being inferred from forum posts.

Status-only application tracking

The application view intentionally keeps editing simple: open the original job or application page, then update the current status from the dropdown. Rejection is a first-class status rather than a hidden archive action.

Application progress with assessment, interview, and rejected states

Recruitment email review

The optional email workflow reads only INBOX through TLS IMAP, classifies minimized recruitment events locally, and places them in a review queue. Nothing changes an application until you confirm that specific proposal. Credentials stay in environment variables and are never written to the workspace.

Personal QQ, NetEase, and Gmail accounts can be detected from the mailbox domain. Tencent, NetEase, and Aliyun enterprise-mail presets are also available, with a custom TLS IMAP fallback. See job-radar/references/email-sync.md for provider configuration and privacy details.

How it works

flowchart LR
    A["Resume or onboarding answers"] --> B["Confirmed privacy-safe profile"]
    B --> C["Current public job sources"]
    C --> D["Normalize, verify, and deduplicate"]
    D --> E["Explainable matching and company grouping"]
    E --> F["Local recommendation dashboard"]
    F -->|"User confirms application"| G["Local application record"]
    G --> H["Assessment / Interview / Offer / Rejected"]
    I["Optional read-only recruitment email sync"] --> J["Local review proposal"]
    J -->|"User confirms"| H
Loading

Codex handles resume interpretation and current web discovery. The bundled Python runtime provides deterministic schemas, source precedence, link classification, deduplication, scoring, storage, history, privacy export, and the local dashboard.

30-second start

1. Install the skill

git clone https://github.com/guiyangyuan/job-radar.git
mkdir -p ~/.codex/skills
cp -R job-radar/job-radar ~/.codex/skills/job-radar

Restart Codex so the new skill is discovered. Job Radar requires Python 3.11 or newer and has no third-party Python dependencies.

2. Build a profile

Attach a resume or answer the guided questions:

Use $job-radar to build my job-search profile from this resume.

Job Radar removes contact details and full resume passages before saving the structured profile. It requires confirmation before recommending roles.

3. Find jobs

Use $job-radar to find current campus roles for AI application,
AI full-stack, Agent, and frontend development in Mainland China.

The result is written to a local workspace such as ./job-radar-data, with source evidence and last-verification timestamps preserved.

4. Open the dashboard

Use $job-radar to open my application dashboard.

Or run the deterministic CLI directly:

python3 ~/.codex/skills/job-radar/scripts/job_radar.py serve \
  --workspace ./job-radar-data

The editable service binds only to 127.0.0.1 and opens a tokenized local URL.

Example prompts

Use $job-radar to refresh my job pool and show only newly verified roles.

I submitted the application for JOB_ID. Record it as applied.

The company invited me to a technical interview. Update the application status.

Export a read-only privacy-safe snapshot of my current progress.

Sync my recruitment emails from the last 60 days and show proposals for review.

Job Radar treats visiting an application page as navigation only. It creates an application record only after the user explicitly confirms submission.

Matching and evidence

The configurable score totals 100 points:

Signal Default What is measured
Role 30 Target role and alias overlap
Skills 30 Required skills supported by visible job evidence
Eligibility 15 Recruitment type, experience, education, and graduation window
Location 10 City and remote-preference match
Preference 10 Preferred industry and company type
Freshness 5 Publication and verification recency

Missing evidence never receives full credit. Expired jobs and strict hard-filter mismatches are filtered rather than quietly ranked lower.

Source precedence is intentionally conservative:

  1. Employer career pages
  2. NCSS and Guopin
  3. Nowcoder, Shixiseng, and public university career centers
  4. Other public link-level discovery sources, followed to an employer page when possible

Search snippets alone never establish that a role is active or that a URL is an exact job-detail page.

Local data model

job-radar-data/
├── profile.json             # confirmed search preferences
├── jobs.json                # normalized and scored job pool
├── applications.json        # user-confirmed application states
├── company-policies.json    # verified quotas and role-change rules
├── email-sync.json          # hashed mailbox identity and IMAP cursor only
├── email-events.json        # minimized, reviewable progress proposals
├── settings.json
├── backups/                 # bounded canonical backups
└── exports/                 # privacy-filtered HTML snapshots

The workspace is separate from the installed skill. The repository ignores job-radar-data/, backups, exports, temporary files, and Python caches.

Privacy and safety boundaries

  • Never stores credentials, cookies, MFA codes, identity numbers, personal email addresses, phone numbers, exact addresses, or full resume text.
  • Never submits an application or claims that one was submitted.
  • Never bypasses login, CAPTCHA, MFA, robots rules, rate limits, or anti-automation controls.
  • Never invents deadlines, eligibility rules, salaries, locations, job IDs, or application URLs.
  • Never advances an application stage without explicit user confirmation or user-provided evidence.
  • Never stores mailbox credentials, message bodies, raw MIME, attachments, or full sender addresses; email access is explicit, read-only, and limited to INBOX.
  • Removes notes, evidence links, and their history copies from privacy exports.
  • Rejects non-loopback dashboard binding, malformed JSON, unsafe URLs, stale writes, and oversized request bodies.

See job-radar/SKILL.md for the complete agent workflow and job-radar/references/source-policy.md for source classification rules.

CLI reference

Command Purpose
init Create canonical local files and bounded backup/export folders
validate Validate versions, fields, URLs, dates, statuses, and score weights
merge-jobs Deduplicate normalized job JSON and apply source precedence
migrate-links Conservatively classify legacy job destinations
merge-policies Merge official company application rules
score Apply hard filters and the explainable six-part score
apply Record a user-confirmed application
update Change an application and append an auditable history event
email-test Verify read-only IMAP configuration without persisting credentials
email-sync Fetch a bounded window and create local review proposals
email-summary Summarize pending email proposals and canonical application states
serve Run the token-protected loopback dashboard
export Generate a self-contained read-only HTML snapshot
summary Summarize job-pool and application states

Run python3 job-radar/scripts/job_radar.py --help for all options.

Repository structure

job-radar/
├── README.md
├── docs/assets/                 # synthetic dashboard screenshots
└── job-radar/
    ├── SKILL.md                    # Codex workflow and boundaries
    ├── agents/openai.yaml          # skill-list metadata
    ├── assets/dashboard-template.html
    ├── references/                 # schemas and source policy
    ├── scripts/job_radar.py        # standard-library CLI
    ├── scripts/job_radar_lib/      # deterministic domain modules
    └── tests/                      # unit and integration coverage

Development

Run the same suite used by GitHub Actions:

PYTHONPATH=job-radar/scripts \
  python3 -m unittest discover -s job-radar/tests -t job-radar -v

The current suite contains 161 tests covering schemas, scoring, source precedence, links, company grouping, application history, email privacy and classification, IMAP behavior, privacy exports, dashboard rendering, and loopback-server security.

License

MIT © 2026 Guiyang Yuan

About

Local-first Codex skill for Mainland China job discovery and application tracking.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages