feat(linkedin): add recommended jobs adapter with GraphQL pagination support

[RickSanchez88E]

Description

Adds a new linkedin recommended command that crawls LinkedIn's personalized job recommendation feed (JYMBII algorithm, at /jobs/collections/recommended/). Unlike the existing linkedin search adapter (REST Voyager API), this endpoint uses GraphQL (/voyager/api/graphql) and requires a browser session.

File: adapters/linkedin/recommended.yaml

Technical Details

• API: LinkedIn uses GraphQL with queryId voyagerJobsDashJobCards.* (version-hashed, discovered dynamically via Performance API)
• Auth: strategy: header with CSRF token extracted from JSESSIONID cookie
• Pagination: Batches of 24 items, automatic multi-page crawl via start offset
• Unlimited mode: --limit 0 crawls until no more items (limit > 0 ? limit - fetched : BATCH loop)
• Easy Apply detection: Checks footerItems[].type === "EASY_APPLY_TEXT" (not easyApplyUrl which doesn't exist in this API)
• Workplace type: Parsed from secondaryDescription.text parentheses, e.g. "London (Hybrid)" → workplace_type: "Hybrid"

Output Columns

rank, title, company, location, workplace_type, salary, posted_time, applicant_count, easy_apply, url

Usage

# Default 200 results
autocli linkedin recommended -f json

# Specify count
autocli linkedin recommended --limit 50 -f json

# Unlimited (crawls all available)
autocli linkedin recommended --limit 0 -f json

# Table format
autocli linkedin recommended --limit 20

# CSV
autocli linkedin recommended --limit 100 -f csv

How to Test

Prerequisites: Chrome must be open with LinkedIn signed in, and the AutoCLI Chrome extension must be installed.

# Quick smoke test (5 results)
autocli linkedin recommended --limit 5

# Verify Easy Apply detection (should see "true" values)
autocli linkedin recommended --limit 24 -f json | grep easy_apply

# Verify pagination (should return exactly 50)
autocli linkedin recommended --limit 50 -f json | python3 -c "import json,sys; d=json.load(sys.stdin); print(f'{len(d)} results')"

# Diagnostic (verify auth & API discovery)
autocli linkedin recommended --limit 3 -f json

Known Quirks / Pitfalls

1. GraphQL variable encoding: LinkedIn requires colons (:) and parentheses to remain raw (not URL-encoded) in GraphQL variables. Full encodeURIComponent causes HTTP 400. The adapter uses a partial-encode-then-decode approach.
2. No total count: The API doesn't return a totalCount field. --limit 0 fetches incrementally until the server returns an empty batch.
3. No applicant_count: Unlike the REST search API, this GraphQL endpoint's jobPostingCard doesn't include applicant count. Column is preserved but always returns "N/A".
4. No easyApplyUrl field: Easy Apply detection uses footerItems type — verified via 200-job crawl with ~30% Easy Apply rate.
5. Dynamic queryId: The GraphQL queryId includes a version hash that may change. The adapter discovers it dynamically via performance.getEntriesByType('resource'), so no hardcoded ID to maintain.
6. Workplace type parsing: Workplace type is embedded in the location string in parentheses. Regex extracts On-site/Hybrid/Remote and strips it from the location field.