The job
You need a LinkedIn profile turned into structured JSON you can drop into your ATS, CRM, or research sheet, with fields like role, company, dates, education, and skills already mapped. Manual copy-paste from collapsed sections is slow, misses lines, and creates messy text that takes longer to clean than to collect.
Why this is hard without Sephir
Without Sephir, you either trust brittle scrapers that fail when LinkedIn changes layout, or you copy raw profile text into a separate chat and rebuild structure by hand. Both routes break context and consistency. You still spend time fixing date ranges, normalizing titles, and forcing strict keys before your internal tooling accepts the payload.
How Sephir does it
- Open the target LinkedIn profile and scroll through key sections so visible text is loaded.
- Open Sephir in the same tab with
Cmd+Shift+S. - Ask for a strict JSON schema with exact fields and null rules for missing data.
- Watch the trace run
extractPageText(current tab)and return typed output, not a prose summary. - Verify the keys in the audit timeline and copy JSON or export the run as JSON.
- Save the successful trace as
/extract-profileso the same extraction pattern runs in one command next time.
The skill behind it
This skill is built for structured extraction, not summarization. It reads one profile tab, applies your schema constraints, and returns machine-ready JSON you can validate before import.
What it costs
Sephir runs this on your own ChatGPT Plus (Codex OAuth) or your own API key. Typical usage is ~4,000–8,000 input tokens and ~500–1,500 output tokens on Claude Opus 4, GPT-5, or Gemini 3 Pro. Single-turn runs fit Free; multi-tab extraction and saved skills are in Pro Lifetime (see ).