nomad-agent-job-scrapers

Nomad Agent Job Data

Nomad Agent publishes an interoperability layer for source-specialized job-search Actors. Each Actor understands its source, while every supported consumer receives the same six-root nomad-agent-job-v1 envelope.

Quick Start

1. Choose the LinkedIn or EURAXESS Actor for the market you need.
2. Start with a bounded search and optional enrichment disabled.
3. Require terminal success, the exact build, and a valid RUN-SUMMARY.
4. Keep canonical records when identity, evidence, or provenance matters.
5. Use the flat projection only for table-oriented destinations.

{
  "keyword": "TypeScript developer",
  "location": "Spain",
  "postedWithin": "past-week",
  "maxItems": 20,
  "translateToEnglish": false,
  "aiEnrichment": false
}

When to Use This Project

• Building a recurring new-job alert without repeat deliveries
• Feeding a job board, search product, database, or internal application
• Giving an AI agent bounded tools for finding and monitoring jobs
• Sending normalized records into n8n, Make, Airtable, or Google Sheets
• Preserving source identity and evidence across multiple job markets

Actor Catalog

Canonical Output Contract

Every canonical job record has exactly six top-level keys:

{
  "schemaVersion": "nomad-agent-job-v1",
  "identity": { "source": "linkedin", "externalId": "…", "url": "…" },
  "data": { "title": "…", "company": { "name": "…" }, "location": { } },
  "custom": null,
  "llm": { "status": "not_requested" },
  "raw": null
}

Source semantics: null means unknown or unavailable. [] means the source established that the field is empty. Integrations must not collapse the distinction.

Deduplicate with jobKey = source:externalId. A title or company name is display data, not job identity.

Integration Packs

Importing a template does not supply credentials or activate a schedule. A named destination must be tested independently before delivery is described as live.

Delivery Checklist

Check all four layers before using a completed run:

• Actor reached a supported terminal status
• The run used the exact expected build
• RUN-SUMMARY validates against the current contract
• Summary delivery count reconciles with the default dataset
• Every record validates as the expected source-specific canonical shape
• At most one bounded retry is honored for a usable partial result

Public Quality Benchmark

The open benchmark scores final customer-facing records, unsupported fills, deterministic-field preservation, provenance integrity, repeated-run completion, and selected-field translation separately for LinkedIn and EURAXESS.

An official accuracy percentage will be published only after blind independent human review and adjudication. Synthetic fixtures and internal positive-heavy sets are not presented as production accuracy.

Security and Responsible Use

• Prefer OAuth and least-privilege MCP tool scopes.
• Never commit Apify tokens, destination credentials, or private datasets.
• Review source terms, licensing, privacy requirements, and applicable law.
• Do not publish personal data without confirming destination access and retention.

This is an independent community project. It is not affiliated with or endorsed by LinkedIn, EURAXESS, the European Commission, or the operators of source websites.

Resources

• Canonical and flat contracts — output identity and semantics
• MCP setup — scoped configuration for supported clients
• Compatibility matrix — current inputs and integration boundaries
• Benchmark contract — public review and scoring artifacts

Start with the Source You Need

Use LinkedIn for broad professional hiring markets or EURAXESS for research and academic opportunities. Both return the same six-root contract, so one consumer can support either source without erasing source-specific evidence.