docs: AI 突出求职 Agent 定位与学习入口

README.en.md

@@ -1,23 +1,43 @@
-# OfferPilot — Open-source AI Job Application Tracker
+# OfferPilot — An Open-source, Local-first Job-search Agent
 
 [简体中文](README.md) | [English](README.en.md)
 
-OfferPilot by offercontext is a local-first workspace for individual
-job seekers. Keep applications, resumes, interviews and offers together,
-with optional AI assistance for preparation and review.
+**Prepare for your next interview—and explore how an agent gets things done.**
 
-## Features
+OfferPilot by offercontext brings applications, resumes, interviews and offers into one workspace.
+Its built-in Pilot agent uses your saved job-search materials to find information, draft content and call tools.
+Changes to important job-search records require your confirmation by default. It does not automatically apply for jobs or contact recruiters.
 
-| Feature | Purpose |
-| --- | --- |
-| Application tracking | Organize companies, roles, application stages and interview schedules |
-| Resume management | Maintain resume versions and review job-specific material suggestions |
-| Mock interviews | Practice with your resume and job description, then review feedback |
-| Interview reviews | Record lessons and select priorities for further practice |
-| Offer comparison | Compare compensation, benefits and response deadlines |
-| Salary negotiation preparation | Prepare questions and communication drafts |
+For AI builders, the same project offers illustrated scenarios, small experiments and source walkthroughs
+covering tool use, context, write approval and execution control.
+
+[Windows unsigned test build](https://github.com/offercontext/offerPilot/releases/tag/v0.1.0-desktop.1) · [Docker / source setup](#getting-started) · [Explore the agent](#use-it-and-learn-how-it-works) · [User guide (Chinese)](docs/product-manual/产品说明书.md)
+
+The Windows installer is an experimental prerelease; read the [test-build limitations](#getting-started) before installing.
+Basic application tracking does not need AI. AI features use your configured model service and may send relevant data off-device and incur provider fees.
+This English README does not imply that every product screen or tutorial is available in English.
+
+## Watch an interview become a saved record
+
+These screenshots come from a real operation captured on October 8, 2026, using a real model and isolated fictional data
+at source revision `c0a447bb`. They are not a full acceptance test of the current Windows release.
+The company, role and date are demonstration data, not a pending task.
 
-New to AI agents? Start with [how an interview gets recorded](docs/learn/01-record-an-interview.md), or browse the [learning path](docs/learn/README.md). These tutorials are in Chinese and require no coding or installation to read.
+The request was to add an online, 60-minute interview for the fictional Yunlan Data test-development role,
+starting at 15:00 Beijing time (UTC+8) on October 15, 2026.
+
+**Find the application → propose the details → you review and confirm → save → verify the calendar record.**
+
+1. Pilot uses a tool to find the application and proposes an interview using the returned information.
+2. The confirmation card shows what would be saved. No interview has been added yet.
+3. After approval, the program saves the interview. The calendar lets you check the company, role and time. No notification was sent to a recruiter.
+
+![Actual confirmation card for the fictional interview: October 15, 15:00, online, 60 minutes; interface in Chinese](docs/learn/images/runtime-20261008/05-interview-pending.jpg)
+
+![Actual calendar record after confirmation: the same fictional interview, 15:00–16:00; interface in Chinese](docs/learn/images/runtime-20261008/08-calendar-verified.jpg)
+
+“Done” is not enough: check the saved record.
+Read the [full walkthrough](docs/learn/01-record-an-interview.md) and [capture conditions and verification scope](docs/learn/images/runtime-20261008/README.md) (Chinese).
 
 ## Getting started
 
@@ -43,15 +63,54 @@ and [user guide in Chinese](docs/product-manual/产品说明书.md).
 
 Basic application tracking does not require AI configuration.
 
-## Privacy and limitations
+## How it helps during your job search
 
-Workspace data is stored locally. AI features send relevant materials
-to your configured model service and may incur provider fees.
+| Feature | Purpose |
+| --- | --- |
+| Pilot agent | Query saved information, draft content and use tools, with confirmation before important record changes |
+| Application tracking | Organize companies, roles, stages and interview schedules |
+| Resume and material preparation | Maintain resume versions and review job-specific suggestions |
+| Mock interviews | Practice with your resume and job description, then review feedback |
+| Interview reviews | Record lessons and choose priorities for further practice |
+| Offer comparison and negotiation preparation | Compare known terms and deadlines, then prepare questions and drafts |
+
+See the [real interface screenshots](README.md#真实界面) and [user guide](docs/product-manual/产品说明书.md) (Chinese).
+This repository describes the locally deployed edition. [offerContext Hub](https://hub.offercontext.cn) is a separate online service;
+its features and data handling follow its own disclosures, and should not be assumed identical or connected.
+
+## Haru and Pilot
+
+Pilot is the built-in agent. Haru is an optional character entry point on wide desktop layouts.
+You can hide Haru and keep using Pilot. Haru is an official Live2D sample character, not an original OfferPilot character;
+see the [third-party notices](README.md#第三方角色与运行时).
+
+## Use it, and learn how it works
+
+You do not need to study agent engineering to use the workspace. AI builders can start with one concrete question:
+**why should confirming the same interview twice still create only one record?**
+
+- **Understand:** follow [one interview being recorded](docs/learn/01-record-an-interview.md), then the [four-article short route](docs/learn/README.md). No installation or coding is required to read it.
+- **Experiment:** try the [Python standard-library experiment](docs/learn/examples/README.md): zero records before confirmation, one after approval, still one after repeated confirmation. It uses a simulated model, needs no model key and does not evaluate real model quality.
+- **Read the implementation:** explore the [Harness source walkthroughs](docs/learn/advanced/README.md), including the agent loop, context, approval, cancellation, recovery and validation.
+
+These learning materials are in Chinese and state their source revision and verification limits.
+Reader trials have not yet been conducted; this is a reading and experimentation route, not a guarantee of mastering a complete system.
+
+## Contribute
+
+[Report an issue](https://github.com/offercontext/offerPilot/issues) or read the [contribution guide](CONTRIBUTING.md)
+before proposing a change. Describe the goal, scope and checks you actually ran.
+
+## Privacy and limitations
 
-OfferPilot does not automatically submit applications or message recruiters.
-Review generated content before using it.
+- Workspace data is stored locally. AI features send relevant materials to your configured model service and may incur provider fees. Local-first does not mean all AI processing stays on your device.
+- Changes to important job-search records require confirmation by default. Review the confirmation card before proceeding.
+- OfferPilot does not automatically submit applications or message recruiters. Review generated content, dates, numbers and commitments before using them.
+- Interview recordings remain in the current page and are not uploaded or persisted. The optional offline Whisper model downloads to the browser cache only after you request it.
+- AI can make mistakes. Decisions about applications, offers and negotiation remain yours.
 
 ## License
 
-See [LICENSE](LICENSE) and the
+[AGPLv3](LICENSE) covers the project, not the Live2D character, model data or runtime.
+Haru and Cubism Core belong to Live2D Inc.; their separate terms and the offline speech-model notices are in the
 [third-party notices](README.md#许可证).