Changelog

All notable changes to ApplyPilot will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[0.2.0] - 2026-02-17

Parallel workers for discovery/enrichment - applypilot run --workers N enables ThreadPoolExecutor-based parallelism for Workday scraping, smart extract, and detail enrichment. Default is sequential (1); power users can scale up.
Apply utility modes - --gen (generate prompt for manual debugging), --mark-applied, --mark-failed, --reset-failed flags on applypilot apply
Dry-run mode - applypilot apply --dry-run fills forms without clicking Submit
5 new tracking columns - agent_id, last_attempted_at, apply_duration_ms, apply_task_id, verification_confidence for better apply-stage observability
Manual ATS detection - manual_ats list in config/sites.yaml skips sites with unsolvable CAPTCHAs (e.g. TCS iBegin)
Qwen3 /no_think optimization - automatically saves tokens when using Qwen models
config.DEFAULTS - centralized dict for magic numbers (min_score, max_apply_attempts, poll_interval, apply_timeout, viewport)

Config YAML not found after install - moved config/ into the package at src/applypilot/config/ so YAML files (employers, sites, searches) ship with pip install
Search config format mismatch - wizard wrote searches: key but discovery code expected queries: with tier support. Aligned wizard output and example config
JobSpy install isolation - removed python-jobspy from package dependencies due to broken numpy==1.26.3 exact pin in jobspy metadata. Installed separately with --no-deps
Scoring batch limit - default limit of 50 silently left jobs unscored across runs. Changed to no limit (scores all pending jobs in one pass)
Missing logging output - added logging.basicConfig(INFO) so per-job progress for scoring, tailoring, and cover letters is visible during pipeline runs

Blocked sites externalized - moved from hardcoded sets in launcher.py to config/sites.yaml under blocked: key
Site base URLs externalized - moved from hardcoded dict in detail.py to config/sites.yaml under base_urls: key
SSO domains externalized - moved from hardcoded list in prompt.py to config/sites.yaml under blocked_sso: key
Prompt improvements - screening context uses target_role from profile, salary section includes currency_conversion_note and dynamic hourly rate examples
acquire_job() fixed - writes agent_id and last_attempted_at to proper columns instead of misusing apply_error
profile.example.json - added currency_conversion_note and target_role fields

6-stage pipeline: discover, enrich, score, tailor, cover letter, apply
Multi-source job discovery: Indeed, LinkedIn, Glassdoor, ZipRecruiter, Google Jobs
Workday employer portal support (46 preconfigured employers)
Direct career site scraping (28 preconfigured sites)
3-tier job description extraction cascade (JSON-LD, CSS selectors, AI fallback)
AI-powered job scoring (1-10 fit scale with rationale)
Resume tailoring with factual preservation (no fabrication)
Cover letter generation per job
Autonomous browser-based application submission via Playwright
Interactive setup wizard (applypilot init)
Cross-platform Chrome/Chromium detection (Windows, macOS, Linux)
Multi-provider LLM support (Gemini, OpenAI, local models via OpenAI-compatible endpoints)
Pipeline stats and HTML results dashboard
YAML-based configuration for employers, career sites, and search queries
Job deduplication across sources
Configurable score threshold filtering
Safety limits for maximum applications per run
Detailed application results logging