Local Resume Autofill

A local-first, privacy-boundary-clear Edge/Chrome resume autofill extension. Resumes are encrypted with a passphrase-derived key, page adapters are auditable declarative JSON, and agents like Codex can only process sanitized page structures and synthetic test data.

[!WARNING] The current version is a developer preview. The Tencent, Alibaba, ByteDance, JD, Meituan, and Pinduoduo adapters are all fixture-verified and have not yet been verified site-by-site on real application accounts. The extension will not upload attachments and will never auto-submit applications.

Core boundaries

• The production extension only requests storage, activeTab, and scripting.

• No <all_urls>, persistent content scripts, telemetry, cloud sync, or remote rules.

• Resumes are encrypted with PBKDF2-SHA-256 (310,000 iterations) and AES-256-GCM.

• The passphrase is not saved; the session key is only placed in chrome.storage.session and becomes invalid after a browser restart.

• Page scripts cannot read the vault via window.postMessage or external messages.

• Authorization is confirmed by exact hostname before the first fill on each site.

• Only the fields planned for use on the current page are sent; the full resume is never handed to the content script.

• After filling, the user reviews and manually submits.

See SECURITY.md and the architecture notes for the full boundaries.

Installing the dev build

Requires Node.js 20+ and Microsoft Edge or Google Chrome.

npm ci
npm run check
npm run build

1. Open edge://extensions or chrome://extensions.

2. Enable "Developer mode".

3. Select "Load unpacked" and point it to dist/extension.

4. Open the extension's "Resume & Permissions", create a vault, and enter your profile data.

Usage

1. Click the extension icon on a job application form page.

2. Click "Preview" to check field references; actual resume values are not shown.

3. Click "Fill". A new hostname will first show an authorization confirmation.

4. Review the page result and handle attachments, CAPTCHAs, and submission manually.

"Capture" exports a *.capture.json that does not contain input values, selected options, attachment names, or screenshots. Capture files may still contain page labels and structure and are excluded from Git by default.

Current adapters

See Adapter authoring for status meanings and the process for adding new adapters.

Connecting Codex or other agents

This project provides a local STDIO MCP server. It can only access the adapters, sanitized captures, and fixtures in the repository; it cannot read the browser vault.

npm run build
Copy-Item .codex/config.toml.example .codex/config.toml
codex mcp list

Project-level MCP configuration must be used in a trusted repository. Codex's STDIO, project-level configuration, and client sharing behavior follow the official OpenAI MCP documentation. Other MCP-capable agents can connect to the same dist/mcp/server.js.

See MCP integration for detailed tools and the confirmation flow.

Verification

npm run typecheck
npm run test:unit
npm run test:integration
npm run test:e2e
npm run audit:extension

Current regression scope:

• 32 unit tests: vault, tamper detection, sanitization, DOM scanning and drivers, 6 site fixtures, MCP paths, and two-phase installation.

• 1 real STDIO MCP integration test: handshake, tool discovery, adapter enumeration, and fixture execution.

• 2 Edge extension end-to-end tests: MV3 loading, scan sanitization, page filling, vault ciphertext, hostname authorization, no submission, and zero external HTTP(S) requests.

• Static audit of the production package: permissions, persistent injection, page message bridge, network APIs, and content-script storage access.

Development

• Architecture and threat model

• Security self-review before v0.3.0 release

• Adapter authoring

• MCP integration

• Contributing

• Changelog

The project is licensed under the MIT License. Third-party job site names are used only for compatibility descriptions; this project has no affiliation with or endorsement from those companies.