feat(jobs): add apply_to_job and save_job tools

[Gabrcodes]

Summary

Two new action tools for the job hunting workflow.

apply_to_job — automates LinkedIn's Easy Apply modal:

• confirm_apply=false (default) dry-runs and reports whether Easy Apply is available
• confirm_apply=true steps through the modal: Next → Review → Submit
• Reports which fields need manual input when required fields are empty
• Scrolls the button into view before clicking (fixes sticky navbar issue from [BUG] connect_with_person fails when Connect button is behind More menu (sticky navbar viewport issue) #304)
• Distinguishes confirmed submission (applied) from unconfirmed (applied_unconfirmed)
• Checks page text when dialog closes unexpectedly to detect silent submissions

save_job — bookmarks a job posting:

• Verifies the button transitioned to "Saved"/"Unsave" after clicking
• Returns save_unavailable if the click didn't take effect

Both are annotated destructiveHint.

Changes

• scraping/extractor.py — apply_to_job, save_job methods
• tools/job.py — apply_to_job, save_job tool registrations

Test plan

• 357 passed, 5 skipped, 0 failures
• ruff check and ruff format pass
• Live test: Easy Apply on a real job posting
• Live test: Save a job and verify in saved jobs list

[greptile-apps]

Greptile Summary

This PR adds apply_to_job and save_job tools that automate LinkedIn's Easy Apply modal and job bookmarking, respectively. All eight issues raised in the prior review round have been addressed: class-name selectors replaced with text-based detection, required-field checks moved before any button click (including Submit), radio-group detection added at the group level, multi-word confirmation phrases used to avoid job-title false positives, and dialog_text prioritised over page_text for the submit confirmation signal.

Confidence Score: 4/5

Safe to merge after considering the worst-case timeout risk; the logic and status reporting are solid.

All prior P0/P1 findings are resolved. One P2 remains: the loop's worst-case runtime (up to ~75 s) cuts close to the 90 s tool timeout, which could leave the dialog open mid-application on slow connections. Reducing max_steps or the per-step dialog timeout would eliminate this risk before merging.

linkedin_mcp_server/scraping/extractor.py — loop timing budget in apply_to_job

Important Files Changed

Filename	Overview
linkedin_mcp_server/scraping/extractor.py	Adds apply_to_job and save_job methods; prior review concerns (class selectors, false-positive text matches, radio group detection, required-field ordering, dialog confirmation scope) are all addressed; one potential timeout risk remains under worst-case step counts.
linkedin_mcp_server/tools/job.py	Registers apply_to_job and save_job tools with destructiveHint/openWorldHint annotations; follows the same pattern as existing tools, error handling and progress reporting are consistent.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[apply_to_job called] --> B[Navigate to job URL]
    B --> C{Easy Apply button found?}
    C -- No --> D{Page text: 'You applied' / 'Application submitted'?}
    D -- Yes --> E[already_applied]
    D -- No --> F[not_easy_apply]
    C -- Yes --> G{confirm_apply?}
    G -- False --> H[confirmation_required]
    G -- True --> I[Click Easy Apply]
    I --> J{Dialog opens within 10s?}
    J -- No --> K[apply_failed]
    J -- Yes --> L[Loop: max 15 steps]
    L --> M{Dialog still open? timeout=3s}
    M -- No --> N{Page text matches submission phrases?}
    N -- Yes --> O[applied_unconfirmed]
    N -- No --> P[applied_unconfirmed dialog_closed_early]
    M -- Yes --> Q{Required fields empty?}
    Q -- Yes --> R[Dismiss + requires_input]
    Q -- No --> S{Submit button?}
    S -- Yes --> T[Click Submit, wait 2s]
    T --> U{dialog_text or page_text matches?}
    U -- Yes --> V[applied]
    U -- No --> W[applied_unconfirmed]
    S -- No --> X{Review / Next / Continue button?}
    X -- Yes --> Y[Click + continue loop]
    X -- No --> Z[apply_failed stuck]
    L -- exhausted --> AA[apply_failed max steps]

Loading

Prompt To Fix All With AI

This is a comment left during a code review.
Path: linkedin_mcp_server/scraping/extractor.py
Line: 1843-1845

Comment:
**Worst-case loop duration approaches the tool timeout**

With `max_steps=15`, each iteration sleeps 1 s then calls `_dialog_is_open(timeout=3000)` (up to 3 s), giving up to 60 s of loop time alone. Adding the 2 s submit sleep, ~10 s modal wait, and page navigation puts the worst case at ~75–80 s against a `TOOL_TIMEOUT_SECONDS` of 90 s. On a slow connection or a long form this will silently time out before the loop finishes, leaving the dialog open mid-application with no cleanup.

Consider either lowering `max_steps` (10 is plenty for any observed LinkedIn Easy Apply form), reducing `_dialog_is_open(timeout=…)` to 1000 ms inside the loop, or adding an explicit elapsed-time guard.

How can I resolve this? If you propose a fix, please make it concise.

_{Reviews (9): Last reviewed commit: "fix: dismiss dialog on unconfirmed submi..." | Re-trigger Greptile}