Linkedin

LinkedIn platform skill (jessy)

DOM extraction goes through the Claude Chrome extension (claude --chrome). You do not need CSS / ARIA selectors — ask the page in natural terms ("the visible job cards", "the description section", "the requirements heading"). This skill captures the semantics of LinkedIn's job pages so you know what to look for, not how to query the DOM.

When this loads

Triggered automatically when the active tab URL contains linkedin.com/jobs/ or linkedin.com/company/ and you are scanning, scoring, or extracting jobs.

URL patterns

A search result page often shows the detail of the currently-selected job inline in a right-side rail; the URL gains currentJobId=<id>. Treat that as a detail view — no need to navigate to the standalone /jobs/view/<id>.

Page semantics — search list

• A scrollable left column containing many job cards.
• Each card has: title, company name, location, posted-time, sometimes an "Easy Apply" badge and a "Promoted" tag.
• Clicking a card loads the detail in the right rail and updates the URL.
• Cards lazy-load on scroll: scroll the list container into view and to the bottom to materialize all cards on the current page.
• Pagination: a numbered pager at the bottom of the list, or next-page button. Some surfaces use infinite scroll instead.

Page semantics — detail (rail or full page)

• Header: title, company link, location, posted date, applicant count.
• Body: a long "About the job" / description block.
• Inside the description, look for requirement headings — see below.
• Sometimes a "Meet the hiring team" / recruiter card; ignore for v1.

Requirement headings (extract → req_hard)

Match any of these heading variants, case-insensitive, anywhere in the description body:

• "Qualifications"
• "Requirements"
• "What we're looking for"
• "Must have"
• "Required skills"
• "Basic qualifications"
• "Minimum qualifications"
• "You have"
• "About you"

The bullets / lines under that heading until the next heading = req_hard.

Nice-to-have headings (extract → req_nice)

• "Nice to have"
• "Bonus"
• "Bonus points"
• "Preferred qualifications"
• "Preferred"
• "Plus"
• "Would be nice"
• "Pluses"

Bullets / lines under those = req_nice. If a single heading mixes both (e.g. "Qualifications" with sub-list "Required" + "Preferred"), split accordingly.

Company info (extract → companies row)

On the company page (linkedin.com/company/<slug>/):

• About / overview section → summary (one short sentence is fine).
• Employee count / company size band → size (e.g. "11-50", "51-200", "10,001+").
• Industry, headquarters available; not stored in v1.

Pagination strategy

For each search tab, walk pages up to platforms.linkedin.max_pages (config; legacy linkedin.max_pages still works for old configs).

Stopping rule (same-list detection): before opening any card on a new page, compare the first 3 card URLs against the first 3 URLs from the previous page. If they match, the pager did not advance — stop walking that tab.

Lazy-load handling

• Cards on the search list lazy-load on scroll. Scroll the list to the bottom before enumerating, otherwise you miss cards.
• Long descriptions sometimes show a "See more" link — expand it before reading, otherwise the requirements section may be truncated.

Boundary / skip rules (read by jessy-scan)

• Title prefilter: drop cards whose title matches any global skip_title_keywords entry before opening.
• Attempt boundary: if a canonical URL exists in Jessy's attempted state, stop lower/older cards for that tab/feed. Do not use LinkedIn viewed. Canonicalize by stripping query params except currentJobId, or use /jobs/view/<id> as the canonical form.

Auth + access

User is expected to be logged in via the chrome session. If a card or detail page redirects to a login wall, surface that to the user and stop — do not attempt to log in.