Configuration Reference

Configuration reference

Settings page

Setting	Default	Meaning
Minimum fit score	60	Roles scored below this are not drafted
Drafts per run	8	The most drafts one daily search writes (best first)
Minimum company size	500	Employees; smaller companies are skipped before scoring. 0 turns it off
Keep unknown sizes	off	Keep companies whose size can't be found
Triage model	claude-haiku-4-5-20251001	First pass over every scraped posting
Scoring model	claude-sonnet-5-5	Scores postings; also classifies Gmail replies
Drafting model	claude-opus-5-5	Writes CVs, cover letters, answers and interview prep
Send notifications / ntfy topic	off / generated	Schedules and Notifications
Check Gmail / address / app password	off	Gmail Tracking

Config file

Everything above, plus secrets, is stored in ~/.config/applydesk/config.json (mode 600, outside the repo). Keys:

Key	Notes
min_score, max_drafts, min_company_size, keep_unknown_size	as above
models.triage, models.score, models.draft	model IDs
notify_enabled, ntfy_topic	phone alerts
gmail_enabled, gmail_user, gmail_app_password	Gmail check
cv_email	optional: an email that must appear in each CV's text layer, checked after compiling
auth	the dashboard login (salted PBKDF2 hash); set with python -m app set-password

Files you edit

File	What
CLAUDE.md	your identity, target roles, location, deal-breakers
.claude/skills/job-application-assistant/01-candidate-profile.md	the facts every draft may use
app/prompts/claim_bank.txt	pre-approved CV bullets
app/store.py → ANSWERS	standard form answers on the dashboard
app/boards.py → BOARDS	company career boards
app/pipeline.py → LINKEDIN_QUERIES, FREEHIRE_QUERIES, AGGREGATORS	search queries and blocklist
app/instahyre.py → QUERIES, YEARS	Instahyre search
app/data/banned_names.txt	names that must never appear in drafts or be read from email
app/prompts/*.md	the prompts for each step

Environment variables

Variable	Use
ANTHROPIC_API_KEY	use an API key instead of a subscription
CLAUDE_CODE_OAUTH_TOKEN	a subscription token from claude setup-token
TZ	timezone for schedules
CLAUDE_CONFIG_DIR	where the Claude login lives (Docker sets it)

Data files (git-ignored)

Path	Contents
job_search_tracker.csv	every application; the dashboard's source of truth
job_scraper/seen_jobs.json	every job already seen, so it isn't scored twice
app/data/runs.json, app/data/runs/*.log	run history (last 300) and logs
app/data/schedules.json	schedules
app/data/replies.json	job-related emails shown on Today
app/data/gmail_state.json, gmail_seen.json, gmail_backfilled	Gmail check progress
app/data/company_sizes.json	company size cache (90 days)