jobplanet-company-profiles

name	jobplanet-company-profiles
description	Use when codex-job-finder company_profiles need JobPlanet ratings, review counts, remaining not_collected rows, legacy imports, or safe public collection/import handling.

jobplanet-company-profiles

Use when the user asks to fill, import, refresh, inspect, or continue JobPlanet rating/review data in company_profiles.

Default DBs

• Current DB: data/headhunter.db
• Legacy DB: /Users/youngchansong/Documents/projects/job-finder/data/lineage-catalog.db

Baseline Workflow

1. Check current status:

sqlite3 -header -column data/headhunter.db "
SELECT jobplanet_status, COUNT(*) AS count
FROM company_profiles
GROUP BY jobplanet_status
ORDER BY jobplanet_status;"

2. Dry-run legacy import:

pnpm jobplanet:profiles -- --dry-run

3. Import exact normalized-company matches from legacy DB:

pnpm jobplanet:profiles

To retry rows already marked not_found, use:

pnpm jobplanet:profiles -- --live --headed --retry-status not_found

4. Verify:

sqlite3 -header -column data/headhunter.db "
SELECT jobplanet_status, COUNT(*) AS count
FROM company_profiles
GROUP BY jobplanet_status
ORDER BY jobplanet_status;"

Remaining Ratings

Use this path after the legacy import leaves not_collected rows.

1. Export remaining companies for the next collection/import step:

mkdir -p results/jobplanet-company-profiles
sqlite3 -header -csv data/headhunter.db "
SELECT normalized_company, company_display
FROM company_profiles
WHERE jobplanet_status = 'not_collected'
ORDER BY normalized_company;" \
> results/jobplanet-company-profiles/remaining-not-collected.csv

2. Prefer deterministic imports before live collection. First verify the legacy/default DB path exists; in the current /Users/mcs environment no lineage-catalog.db was found at the historical default paths, so skip legacy import unless a trusted DB/CSV is explicitly present. If another trusted DB/CSV exists, import only rows whose normalized_company exactly matches company_profiles.normalized_company.

3. If implementing a live collector, target only the exported not_collected rows and upsert every attempted company as one of:

   • ok: rating/review data found for an exact JobPlanet match.
   • not_found: public search completed but no candidate matched.
   • blocked: JobPlanet returned CAPTCHA/authwall/Cloudflare/rate-limit/login-wall/HTTP 403/429.
   • failed: transient or parser failure worth retrying later.

4. For live searches, use the primary company name before a parenthesized alias: xxx(yyy) must search JobPlanet as xxx, while keeping the DB key as the original normalized_company.

5. Do not automatically import partial/fuzzy matches into jobplanet_rating or jobplanet_review_count. Store partial evidence only in metadata or a separate review artifact unless the user explicitly approves a manual match.

Existing Implementation To Reuse

Use the old job-finder implementation as a reference, not as an unsafe execution recipe:

• /Users/youngchansong/Documents/projects/job-finder/src/services/company-enrichment.ts
  • Reuse parser ideas: parseRating(), parseReviewCount(), search candidate extraction, exact vs partial candidate choice.
  • Reuse status handling shape: ok, not_found, blocked, failed.
• /Users/youngchansong/Documents/projects/job-finder/src/jobplanet-scores/cli.ts
  • Reuse target selection, dry-run behavior, and status-preserving upsert pattern.
• /Users/youngchansong/Documents/projects/job-finder/.claude/skills/jobplanet-scores/SKILL.md
  • Treat as historical context only. Do not copy authenticated browser/CDP instructions into this repo.

When porting code, adapt it to SearchJobsRepo and company_profiles; keep pnpm jobplanet:profiles as the primary CLI surface or add subcommands under the same command.

Policy

• Prefer deterministic DB import over live scraping.
• Import only exact normalized_company matches.
• Do not import partial/fuzzy matches into rating fields automatically.
• Public unauthenticated HTTP/browser access is allowed only while pages are normally accessible.
• Do not bypass CAPTCHA, authwall, MFA, OAuth/risk prompts, Cloudflare, rate limits, or login walls.
• Do not use cookies, browser profiles, local/session storage, Bitwarden vault data, secrets, or authenticated CDP sessions.
• If JobPlanet blocks access, store blocked or report it; do not invent data.

Acceptance

• Status counts are checked before and after the run.
• All attempted companies are represented in company_profiles.
• ok rows have exact-match provenance, rating/review count when present, URL, observed timestamp, and raw metadata.
• A remaining CSV or run artifact is saved under results/jobplanet-company-profiles/ when unfinished rows remain.