JOSINT - the open-source job intelligence system

JOSINT monitors public recruiter job boards, normalizes job advertisements, assigns controlled industry/function labels, and can infer likely hidden employers behind recruiter postings.

It can run locally as a Codex skill or on a server/OpenClaw schedule. Runtime data, reports, state, and secrets are kept outside the source code and resolved from relative workspace paths.

What It Does

• Crawls enabled recruiter job boards for a date range or the previous business day.
• Stores clean job records in SQLite.
• Keeps job title and JD text in separate fields.
• Labels jobs using fixed industry and function enums.
• Uses MiniMax-M3 plus optional Metaso search to infer likely hidden employers.
• Generates Chinese Markdown reports.
• Optionally syncs results to Feishu Bitable and sends Feishu IM summaries.
• In scheduled Feishu mode, only jobs newly created in Bitable are sent to employer inference; re-seen or updated jobs are deduplicated without re-analysis.

Quick Start

cp skills/web-ad-radar/.env.example .env
python skills/web-ad-radar/scripts/run_radar.py --workspace . --crawl-only
python skills/web-ad-radar/scripts/run_radar.py --workspace . --offline-sample --companies hays --crawl-only

Scheduled/OpenClaw-style run:

python skills/web-ad-radar/scripts/radar_cron.py --workspace .

Configuration

Copy skills/web-ad-radar/.env.example to .env and fill only the integrations you need.

• MiniMax text model is used for extraction/fallback labeling where configured.
• MiniMax-M3 is used for hidden-employer reasoning.
• Metaso is used for evidence search and candidate-employer verification.
• Feishu variables are only required for Bitable/IM sync mode.
• FEISHU_ANALYSIS_LOG_TABLE_ID is optional. When configured, each inference result is also appended to a separate analysis log table with reasoning, searches, evidence, and review flags.

Outputs

• reports/josint-YYYY-MM-DD.md
• data/jobs.sqlite
• data/api_usage.jsonl
• data/bitable_state.json when Feishu sync is enabled

Feishu Data Contract

The main Bitable keeps one canonical row per job advertisement, deduplicated by URL and normalized url_hash.

• month is set only when a row is first created and is intended for monthly views.
• last_seen_month is refreshed whenever a row is seen again.
• analysis_*, reasoning_summary, and *_json fields store the latest hidden-employer inference state.
• A separate optional analysis log table can be enabled with FEISHU_ANALYSIS_LOG_TABLE_ID to preserve the inference trail for every run.

Documentation

• Skill SPEC
• Source Registry
• Employer Inference Playbook

Tests

cd skills/web-ad-radar/scripts
python -m unittest discover -s ../tests -v

License

Apache License 2.0. See LICENSE.