agentleFS
farukzahrafarukzahra/agent-skills/skills/find-job/SKILL.md

find-job

Skill0 starsChanged 2 months agoMITView on GitHub

Search LinkedIn Jobs in the Cursor browser using the user's resume, verify true remote eligibility from Brazil/LATAM, apply worldwide remote filters, flag suspicious employers, and submit Easy Apply automatically. Use when the user invokes /find-job or asks to find and apply to remote jobs on LinkedIn.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

Show its rules and text

We split this skill into the rules it gives an agent

21 rules. About 0 are common, 16 are distinctive, and 5 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈16

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 5

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    If the user did not attach or name a resume path in the same message:

    line 35
  2. Distinctive:

    Ask: *Where is your resume?

    line 37
  3. Distinctive:

    If they say default or give no path after asking, load Faruk's default:

    line 38
  4. Very specific:

    Read the file with Read. For PDF, extract text from readable content.

    line 46
  5. Distinctive:

    If not specified, ask once with defaults below.

    line 58
  6. Distinctive:

    Do not ask for their password.

    line 85
  7. Very specific:

    After confirmation, browser_snapshot again and verify feed/search UI (not login form).

    line 86
  8. Distinctive:

    Default: only analyze jobs where apply is on LinkedIn (Easy Apply).

    line 105
  9. Distinctive:

    When collecting results, prefer jobs whose action button reads Easy Apply, Apply (LinkedIn modal), or Candidatura simplificada — not Apply …

    line 107
  10. Distinctive:

    Run remote geo check (§5) — before scoring for apply

    line 113
  11. Distinctive:

    Stop early if LinkedIn shows captcha, rate limit, or repeated failures — report and switch to manual review of collected links.

    line 116
  12. Distinctive:

    Read location on card + full description.

    line 126
  13. Very specific:

    Hard rule: if geo class is REMOTE_REGION_LOCKED, REMOTE_UNCLEAR, or NOT_REMOTE → tier Skip regardless of skills score.

    line 152
  14. Distinctive:

    Must be US citizen / no sponsorship

    line 156
  15. Distinctive:

    Never pay anything. Warn if posting asks for upfront payment or crypto.

    line 172
  16. Very specific:

    If login expired, try credentials from ../secrets/linkedin/farukz.txt only after user stored them there.

    line 212
  17. Distinctive:

    If user asks, prepare:

    line 237
  18. Distinctive:

    Do not store LinkedIn session cookies in repo files.

    line 243
  19. Very specific:

    Prefer browser_snapshot over blind clicks

    line 250
  20. Distinctive:

    When user says default or does not specify a path after you ask:

    line 257
  21. Distinctive:

    Commit LinkedIn cookies, session tokens, or passwords

    line 280

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

/find-job — LinkedIn job search from resume

Find remote, international roles on LinkedIn Jobs (work from Brazil — not jobs that require living or working in the USA). Prefer USD when listed. The user logs in manually; the agent drives search, filtering, scoring, Easy Apply submission, and reporting.

Every /find-job run must attempt applications on eligible Easy Apply jobs (see §7), not search-only.

When to use

  • User invokes /find-job
  • User asks to search LinkedIn for remote international jobs from Brazil/LATAM (not US-residency, not Dubai/UAE-only, etc.)
  • User wants help avoiding suspicious or fake job postings

Hard limits (tell the user upfront if relevant)

  • Login is manual — LinkedIn blocks automated sign-in; pause until the user confirms they are logged in.
  • USD salary is not guaranteed — LinkedIn often shows estimates; treat salary as a signal, not proof.
  • Fake-company detection is heuristic — score risk; never claim 100% certainty.
  • Fragile automation — LinkedIn UI changes and anti-bot measures can break steps; prefer small batches (first 1–2 result pages).
  • Personal use only — do not scrape at high volume or store LinkedIn credentials.

Workflow

1. Resolve resume input

If the user did not attach or name a resume path in the same message:

  1. Ask: Where is your resume? (paste text, file path, or say "default")
  2. If they say default or give no path after asking, load Faruk's default:
PriorityPath (relative to workspace)Notes
1../faruk/frontend/public/assets/Faruk Zahra - CV - Resume.pdfGenerated PDF
2../faruk/frontend/src/views/ResumeView.vueStructured source (skills, summary, titles)
3../faruk/frontend/dist/index.html + /resume routeOnly if built site is needed

Read the file with Read. For PDF, extract text from readable content; if poor, fall back to ResumeView.vue.

Build a resume profile (keep in memory for this session):

  • Target titles (e.g. Senior Fullstack Engineer, Java Architect)
  • Top skills / keywords (10–20)
  • Years of experience, seniority band
  • Languages (English level matters for US roles)
  • Locations to avoid: US-only remote, must live in US, US work authorization required (see reference/candidate-geo-defaults.md)

2. Confirm search parameters

If not specified, ask once with defaults below. User may answer in Portuguese.

ParameterDefault (Faruk-style)Options
Work modeRemoteRemote, Hybrid (only if user asks)
GeographyWorldwide international remote (from Brazil)Worldwide, LATAM, Global — not US-residency jobs
US work authorizationNo (Faruk default)User may override
CompensationPrefer USD when listedUSD, EUR, any
SeniorityFrom resumeEntry, Mid, Senior, Staff
Apply channelEasy Apply onlyEasy Apply, any
KeywordsFrom resume titles + top skillsUser override
Max jobs to analyze2510–50
Max Easy Apply submits20 per run1–20

Record the final parameters before opening the browser.

3. Open LinkedIn Jobs (Cursor browser)

Use cursor-ide-browser MCP:

  1. browser_tabs → list tabs
  2. browser_navigate → https://www.linkedin.com/jobs/
  3. browser_snapshot → check login state

If not logged in:

  • Tell the user: Please log in to LinkedIn in this tab, then reply "logged in".
  • Do not ask for their password.
  • After confirmation, browser_snapshot again and verify feed/search UI (not login form).

4. Run search with filters

Build a search URL or use UI filters. Prefer URL for repeatability:

https://www.linkedin.com/jobs/search/?keywords={encoded_keywords}%20international%20worldwide&location=Worldwide&f_WT=2&f_AL=true&f_TPR=r604800

Geo rules and search defaults: reference/candidate-geo-defaults.md. Do not default to location=United States.

FilterLinkedIn UI (EN)LinkedIn UI (PT-BR)URL param (when stable)
RemoteRemoteRemotof_WT=2
Easy ApplyEasy ApplyCandidatura simplificadaf_AL=true
Date postedPast weekÚltima semanaf_TPR=r604800
ExperienceSenior / Director…f_E=4,5,6 (verify in UI)

Default: only analyze jobs where apply is on LinkedIn (Easy Apply). Skip listings whose primary CTA is Apply on company website / Candidatura no site da empresa unless the user explicitly opts out.

When collecting results, prefer jobs whose action button reads Easy Apply, Apply (LinkedIn modal), or Candidatura simplificada — not Apply on company website.

For each results page (max 2 pages unless user asks for more):

  1. browser_snapshot → collect job cards (title, company, location line, link)
  2. Open each candidate → read full description
  3. Run remote geo check (§5) — before scoring for apply
  4. Append only jobs that pass geo gate (or mark Skip with evidence)

Stop early if LinkedIn shows captcha, rate limit, or repeated failures — report and switch to manual review of collected links.

5. Remote geo verification (required before apply)

Many listings say Remote but mean remote only from the US, Dubai/UAE, EU, etc. — not from Brazil.

Full rules: reference/remote-geo-check.md.

For every job, before Easy Apply:

  1. Read location on card + full description.
  2. Classify: REMOTE_BRAZIL_OK | REMOTE_LATAM_OK | REMOTE_UNCLEAR | REMOTE_REGION_LOCKED | NOT_REMOTE.
  3. Only REMOTE_BRAZIL_OK or REMOTE_LATAM_OK may proceed to apply.
  4. Log classification + short evidence quote in the report.

Skip apply examples:

  • Remote Dubai / UAE / GCC only
  • Remote US only + US work authorization
  • Hybrid or on-site
  • “Remote” with no geo policy → REMOTE_UNCLEAR (do not guess)

6. Score each job against the resume

For each job, compute:

DimensionWeightNotes
Title / seniority match25%Align with resume titles
Skills overlap30%Stack in description vs resume
Remote / geo fit30%Must be REMOTE_BRAZIL_OK or REMOTE_LATAM_OK — see §5
Compensation signal10%USD mentioned, salary range if present
Company trust5%From fake-company heuristics (below)

Output tiers: Strong match, Possible, Weak, Skip.

Hard rule: if geo class is REMOTE_REGION_LOCKED, REMOTE_UNCLEAR, or NOT_REMOTE → tier Skip regardless of skills score.

Penalize heavily:

  • Must be US citizen / no sponsorship
  • On-site or hybrid required
  • Clear timezone lock incompatible with Brazil
  • High fake-company risk

See reference/fake-company-heuristics.md and reference/remote-geo-check.md.

7. Fake / suspicious company check

Apply heuristics from reference/fake-company-heuristics.md. For each employer, assign:

  • Risk: low / medium / high
  • 1–3 bullet reasons

Optional: open company LinkedIn page in browser and note employee count, founded date, website link — only when high-value candidates need verification.

Never pay anything. Warn if posting asks for upfront payment or crypto.

8. Easy Apply — automatic submission (required)

On every /find-job execution, after §5 geo check + §6 scoring + §7 fake-company check, submit applications for eligible jobs.

Full browser steps: reference/easy-apply-workflow.md.

Apply when all are true:

  • Geo class REMOTE_BRAZIL_OK or REMOTE_LATAM_OK (§5 — verified in description)
  • Listing is Easy Apply / Candidatura simplificada (f_AL=true results only)
  • Match tier Strong match, or Possible with company risk low
  • Company risk is not high
  • Screening questions can be answered truthfully from the resume (see reference/screening-answers.md)

Screening answers (required):

  1. Snapshot each question text in the Easy Apply modal.
  2. Using the loaded CV, write what the candidate would answer — you are the model; no undefined, no empty required fields, no invented facts.
  3. Fill with browser_type or CDP; confirm value visible before Continue.
  4. Skip job only when honesty requires info absent from CV (work auth, clearance, salary not provided).

Skip apply (log reason in report) when:

  • REMOTE_REGION_LOCKED — e.g. remote US/Dubai/UAE/EU only
  • REMOTE_UNCLEAR — “Remote” without Brazil/LATAM/worldwide evidence
  • NOT_REMOTE — hybrid or on-site
  • External / company-website apply only
  • High fake-company risk
  • US-only / no sponsorship if candidate is Brazil-based and posting excludes international
  • Required question cannot be answered honestly
  • Per-run submit cap reached (default 20 — stop applying and finish report)
  • LinkedIn blocks with captcha or verification

Flow:

  1. Sort candidates: Strong match first, then Possible (low risk).
  2. For each job until cap reached (max 20, then stop): open → §5 geo check → Easy Apply modal → contact → resume → screening → Submit.
  3. Track applied | skipped + reason per job.
  4. If login expired, try credentials from ../secrets/linkedin/farukz.txt only after user stored them there; otherwise pause for manual login.

9. Deliver report

Produce a self-contained HTML file (dark mode, inline CSS — same quality bar as recap):

Save to:

  • docs/job-search/YYYY-MM-DD-find-job-<slug>.html when docs/ exists
  • otherwise find-job-report.html at workspace root

Include:

  1. Search parameters used
  2. Resume source path (no secrets)
  3. Summary counts (found / strong / possible / skipped / high-risk companies)
  4. Table or cards: rank, title, company, location, geo class, match tier, risk, LinkedIn URL, apply status, 2-line rationale
  5. Section Applications submitted this run (applied vs skipped + reasons)
  6. Section High-risk postings to avoid
  7. Section Recommended next steps (profile tweaks, manual follow-ups)

Reply with file path, apply summary (e.g. 3 applied, 2 skipped), and top matches in markdown.

10. Optional follow-up

If user asks, prepare:

  • Short LinkedIn Easy Apply talking points from resume
  • Cover letter snippet for top match
  • Saved search keywords for next session

Do not store LinkedIn session cookies in repo files.

Browser discipline

Follow cursor-ide-browser rules:

  • browser_navigate → browser_lock → actions → browser_lock unlock
  • Prefer browser_snapshot over blind clicks
  • Max 4 failed actions per step; then stop and report blocker

Detailed LinkedIn steps: reference/linkedin-workflow.md.

Default resume (Faruk)

When user says default or does not specify a path after you ask:

  • PDF: ../faruk/frontend/public/assets/Faruk Zahra - CV - Resume.pdf
  • Source: ../faruk/frontend/src/views/ResumeView.vue

Details: reference/default-resume-paths.md.

Project slash command

Consumer repos should add .cursor/commands/find-job.md:

# /find-job

User invoked **`/find-job`**. Read and follow the **`find-job`** skill
(`~/.cursor/skills/find-job/SKILL.md` or `.agents/skills/find-job/SKILL.md`).

If no resume path was given, ask where the CV is; if unspecified, use the default from `../faruk`.
Open LinkedIn in the Cursor browser; user logs in manually; **submit Easy Apply for eligible matches**; deliver ranked HTML report under `docs/job-search/`.

Do not

  • Commit LinkedIn cookies, session tokens, or passwords
  • Claim guaranteed USD salary or visa sponsorship
  • Lie on Easy Apply screening questions (work auth, clearance, etc.)
  • Submit undefined, empty, or fabricated screening answers — use CV-grounded drafts per reference/screening-answers.md
  • Apply to high-risk fake postings or external-only listings
  • Scrape more than needed for the report (default cap: 25 jobs analyzed)

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.