Overview
Open Settings and choose a section (Content generation, Resume scoring, Chat & refinement, Role templates, Job search, and so on) to edit system prompts, models, and resume input formats. Most changes need Save settings. Theme and multi-window apply immediately.
Guidance also exists outside the global system prompts:
- • Role-specific prompt — per role on the Roles page; appended during content generation.
- • Template insights additional instructions — saved from Roles → Analysis (not from Settings → Role templates). Settings holds the model and system prompt only.
- • Your messages — generation instructions, refinement requests, Application Chat questions, and per-run extras you type.
For the full Settings map and multi-window behavior, see Configure the app and Single vs multi-window.
How instructions stack
Claude typically sees layers of guidance in this order:
- 1. Global system prompt for the feature (Settings).
- 2. Role-specific prompt (Roles) — content generation.
- 3. Generation instructions and optional job notes on the application.
- 4. Per-run additional instructions — Improve with AI, interview cheat sheet, template insights.
- 5. Your chat or refine message for that turn.
Models
Every AI feature uses a Claude model from your Anthropic account. In Settings you pick:
- • Default Claude model (Settings → Files & defaults) — fallback for any prompt that does not specify its own model.
- • Per-prompt model — optional override on each feature below. Leave as “Use default model” to inherit the default.
Improve with AI, interview cheat sheet, and template insights also let you override the model for a single run in their modals/panels without changing the saved Settings default. Use Settings → API & updates → Review configured models if Anthropic retires a model ID you still have saved.
Resume input format
Content generation, resume scoring, Application Chat, and section refinement each have a Resume input format picker in Settings. This controls how your resume is sent to Claude (separate from system prompts):
- • Structured text (default) — section IDs and plain text. Lowest API usage.
- • Full HTML — rendered HTML with CSS inlined. More context; still much less than PDF.
- • PDF document — composes and attaches a PDF. Highest usage; best when layout judgment matters. Avoid for Application Chat unless you need it — chat resends resume context each turn.
Typical 1-page resume (excluding the job description): structured text ~600–950 input tokens; HTML ~860–1,200; PDF ~1,850–3,350 per API call.
Content generation
Settings → Content generation · model · system prompt · resume input format
Guides Claude when tailoring resume and cover letter sections for a job. Used by New Application, batch generate from Search, and regenerate on Applications.
- • Enforces factual accuracy and formatting rules (bullets, job headers, section IDs).
- • Works with the role template, base resume content, job description, role-specific prompt, and optional generation instructions / notes.
- • Output is structured JSON mapped to your template sections.
Resume evaluation
Settings → Resume scoring · Resume evaluation model · system prompt · input format
Controls how Claude scores a resume against a job description. Used for baseline (template as-is) and customized evaluations in the wizard, Applications, Search bulk score, and optional “score after generation.”
- • Returns a 0–100 score plus strengths, gaps, and recommendations.
- • Customize to emphasize skills you care about, ATS keywords, or feedback style.
Application chat
Settings → Chat & refinement · Application chat model · system prompt · input format
System instructions for Application Chat — drafting answers to employer questions using the saved resume, cover letter, and job description.
- • Default rules require first-person, factual answers grounded in your materials.
- • Each question you type is a separate user message; resume context is resent each turn — prefer structured text input format for cost.
Content refinement
Settings → Chat & refinement · Content refinement model · system prompt · input format
Used when you refine a single resume or cover letter section — for example “make this bullet more metrics-driven” or “shorten to three lines.”
- • Claude returns JSON with proposed section text and a short summary of changes.
- • Includes job description, base resume, and current section text as context.
Evaluation enhancement
Settings → Resume scoring · Evaluation enhancement model · system prompt
Powers Improve with AI — turning evaluation gaps and recommendations into proposed section updates after you have scored an application.
- • Only suggests edits where a meaningful improvement exists; stays faithful to your experience.
- • Optional per-run model and additional instructions in the Improve with AI modal.
Interview cheat sheet
Settings → Resume scoring · Interview cheat sheet model · system prompt
Powers Generate interview cheat sheet on Applications and in the New Application review step — interview prep from evaluation feedback.
- • Returns overview, strengths to emphasize, gap responses, talking points, and questions to ask.
- • Optional per-run model and instructions in the modal; results are saved with the application.
Template improvement
Settings → Role templates · model · Template Improvement System Prompt · additional instructions on Roles → Analysis
Drives Template insights on the Roles page — analyzing evaluation feedback across applications for one role and recommending template-level fixes (section defaults, role prompt additions, layout notes).
- • System prompt (Settings → Role templates) — overall analysis rules and output structure.
- • Additional instructions (Roles → Analysis, saved from that panel) — extra guidance applied on subsequent runs, e.g. “focus on summary” or “prioritize ATS keywords.”
- • Per-run model & instructions — override for one analysis without changing saved defaults.
JobWeb fit screening
Settings → Job search · Posting fit screening · model · system prompt
When enabled, Claude triages fetched JobWeb listings against your template/role so you can filter high / medium / low fit before generating applications. Per saved search you can turn fit screening on or off independently of the global default.
- • Runs during fetch (and when you re-evaluate fit) — not the same as full resume evaluation scoring.
- • Also configure JobWeb base URL, optional auth token, and default role in the same Settings section.
Role-specific prompt
Roles → select a role → Role-specific prompt
Not a global system prompt — each role has its own instructions appended when generating or regenerating application content for that role. Use it to steer emphasis without rewriting the global content generation prompt.
Template insights may recommend additions to this prompt based on patterns in your evaluation history.
Your own messages
These are not system prompts in Settings, but they shape AI behavior alongside them:
- • Generation instructions — New Application Generate step or application detail; passed into content generation for that posting.
- • Job details notes — optional notes on New Application or the application; extra context for generation.
- • Section refinement requests — the instruction you type when refining a section.
- • Application Chat messages — each employer-specific question you ask.
- • Improve with AI / interview cheat sheet / template insights instructions — optional per-run extras in those modals or the Roles Analysis panel.
Want the full walkthrough?
Installation, multi-window, Settings, search, export, and tracking.
Read the documentation