Each mode file defines one workflow (evaluate, apply, scan, ...). The agent reads the mode plus the shared context and your user files, then executes it. Routing — which user request triggers which mode — lives in the Skill Modes table in AGENTS.md (mirrored in CLAUDE.md).

Mode catalog

File	Mode	Purpose
`oferta.md`	`job`	Full A-G evaluation of a single offer
`ofertas.md`	`jobs`	Multi-job comparison
`auto-pipeline.md`	auto	Full automatic pipeline (evaluate + PDF + tracker) on a pasted JD/URL
`pipeline.md`	`pipeline`	Process the URL inbox (`data/pipeline.md`)
`scan.md`	`scan`	Portal scanner (job discovery)
`batch.md`	`batch`	Mass processing with headless workers
`apply.md`	`apply`	Live application assistant (form filling; never submits)
`pdf.md`	`pdf`	ATS-optimized PDF generation
`latex.md`	`latex`	LaTeX/Overleaf CV export
`cover.md`	`cover`	Cover letter generator
`email.md`	`email`	Application email drafts (draft-only)
`contacto.md`	`contacto`	LinkedIn outreach messages
`deep.md`	`deep`	Deep company-research prompt
`interview.md`	`interview`	Interactive profile & CV onboarding
`interview-prep.md`	`interview-prep`	Company-specific interview intelligence
`interview-redflag.md`	`interview-redflag`	Company red-flag detector
`offer-prep.md`	`offer-prep`	Contract reading companion (offer stage)
`followup.md`	`followup`	Follow-up cadence tracker
`reply-watch.md`	`reply-watch`	Classify employer replies, reconcile tracker
`tracker.md`	`tracker`	Applications tracker overview
`patterns.md`	`patterns`	Rejection pattern detector
`titles.md`	`titles`	Adjacent job-title suggestions
`training.md`	`training`	Training & course evaluation
`project.md`	`project`	Portfolio project evaluation
`add.md`	`add`	Add a project, paper, or role to the CV (confirm-before-write)
`agent-inbox.md`	`agent-inbox`	Queue requests for the next session
`update.md`	`update`	Interactive system update

Shared context and user customization

File	Role
`_shared.md`	System context shared across modes: scoring system, global rules, source-of-truth boundary. System-owned — never put personal data here
`_profile.template.md`	Seed for your `modes/_profile.md` (archetypes, narrative, negotiation scripts)
`_custom.template.md`	Seed for your `modes/_custom.md` (house rules, procedural preferences)

Your copies (_profile.md, _custom.md) are user-layer files: gitignored and never touched by update-system.mjs (see DATA_CONTRACT.md).

Subdirectories

Dir	Contents
`interview/`	Reusable interview skills: prep planner, practice interviewer, post-interview debrief (see interview/README.md)
`heuristics/`	Shared candidate-facing writing heuristics loaded by other modes — `recruiter-side.md` governs PDF summaries, bullets, cover letters, form answers, and outreach
`regional/`	Market calibration modes — `eu-swe.md` calibrates applications for European SWE roles (advisory only)
`ar/ da/ de/ es/ fr/ hi/ id/ it/ ja/ ko/ pl/ pt/ ru/ tr/ ua/ zh/`	Language modes: native translations of the core modes with market-specific vocabulary; each has its own README

Conventions

One file = one mode; the h1 is # Mode: <name> — <purpose>.
Underscore prefix = shared context or template, not a routable mode.
Language selection: explicit user request or language.modes_dir in config/profile.yml wins over JD-language detection (see AGENTS.md).
Mode files are system-layer: edits belong upstream. User personalization goes in _profile.md / _custom.md, never in the mode files themselves.