apply-easy-apply

apply-easy-apply

You drive the LinkedIn Easy Apply flow using claude-in-chrome tools.

Inputs (already in your subagent context)

• The current tab is at the job posting URL.
• cv_pdf_path: path to the prepared PDF (from compose-cv).
• profile_text: profile.md.
• job_id, query_id.
• dry_run: boolean — when true, take a screenshot at the review step instead of submitting.

Pipeline

1. Open the Easy Apply modal

• Find the "Easy Apply" button (look for button with text "Easy Apply" or aria-label containing "Easy Apply").
• Click it.
• Wait for the modal <div role="dialog"> to be present (poll up to 10s).
• If the modal does not open within 10s, retry once (close any overlay, click again). If still failing → return error_modal.

2. Step through the modal

The modal may have 1 to ~5 steps. The "Next" / "Continue" / "Review" / "Submit application" button at the bottom advances. Strategy: at each step, identify all fields, fill them, then click the advance button. Repeat until "Submit application" is the active button.

For each step:

1. Read the page (use mcp__claude-in-chrome__get_page_text or read_page).
2. Identify all input fields in the modal:
   • text inputs (text, email, tel, number)
   • textareas
   • selects / dropdowns
   • radio groups
   • checkboxes
   • file uploads
3. For each field:
   • Pre-filled by LinkedIn (name, email, phone): leave alone.
   • File upload (resume): upload cv_pdf_path via mcp__claude-in-chrome__file_upload.
   • Other field: invoke skill compose-screening-answer with the question text and widget metadata. Type/select the returned answer.
   • If the answer is unanswerable: true AND the field is required: close the modal (Esc or X button) and return error_unanswerable_required.
4. Click the advance button.

3. Detect the final step

The final step shows "Submit application" (not "Next" or "Continue"). Sometimes there's a review screen first; click through it.

4. Detect login wall / CAPTCHA at any moment

If at any point:

• The page contains <input type="password"> or the URL contains /login → return error_login_wall.
• The page contains iframe[src*="captcha"] or a "verify you are human" text → return error_captcha.

Return immediately without retrying.

5. Submit (unless dry-run)

If --dry-run is active (the orchestrator passes a flag in the prompt):

• Take a screenshot of the modal at the review step.
• Save to cvs/<job_id>-dryrun.png via mcp__claude-in-chrome__get_screenshot.
• Do NOT click Submit. Return dryrun_completed.

Otherwise:

• Click "Submit application".
• Wait up to 10s for the confirmation ("Application sent" or similar text, OR the modal closes and a "Done" / "View applied jobs" appears).
• If confirmed → status submitted.
• If not confirmed within 10s → status error_submit_unconfirmed.

6. Return result

Return a structured result to the caller (the job-applier subagent), including:

• status: one of submitted, dryrun_completed, or an error code.
• screening_answers: list of {question, answer} pairs collected during the modal steps.
• cover_letter_excerpt: string or null (Easy Apply rarely has a cover letter field).
• cv_path: the cv_pdf_path you uploaded.

The subagent will log to applied.jsonl.

Pacing

Inside this skill: do NOT add extra waits beyond the modal-loading polls. The orchestrator handles 30–90s pacing between candidates.

Notes

• Some Easy Apply forms have a "Follow company" checkbox. Always uncheck it unless the profile explicitly says otherwise — we don't follow companies just from applying.
• Some forms have a "Save and apply later" option — ignore it.