Jobs Express is a small local app maintained by Datafying Tech. It handles personal information (your profile and resume) and drives a browser you are signed in to, so we take reports seriously and will answer every one.
Please do not open a public issue for a security problem.
Use GitHub's private reporting instead:
- Go to the repository's Security tab.
- Click Report a vulnerability.
- Tell us what you found, how to reproduce it, and what an attacker could do with it.
Private vulnerability reporting is enabled on this repository, so the Report a vulnerability button on the Security tab is the way to reach us. Please include the app version (shown in the sidebar), your operating system, and your Node.js version.
What to expect. We aim to acknowledge a report within 3 business days and to have a fix or a clear plan within 30 days. We will credit you in the release notes unless you would rather we did not. This is a free project with no bug bounty.
In scope:
- Anything that lets a web page or another local user reach the app's API or read the data folder (DNS rebinding, CSRF,
path traversal, a bypass of the Host, Origin,
X-CJAor per-launch token checks). - Anything that lets text from a profile, a job posting or a web page become a command (including the scheduled task, the install and sign-in buttons, and the log helper).
- Writing files outside the data folder, including through resume uploads or profile fields.
- Command injection through profile text, filenames, or the
tools/log.mjshelper. - Ways to make an agent run a tool outside its allow-list, or to widen that list.
- Anything that sends your profile, resume, or logs anywhere other than Claude Code and the sites you apply to.
- Prompt-injection paths that make an agent break a hard stop (create an account, enter a password, click a CAPTCHA, message someone) from the contents of a web page.
Out of scope:
- An attacker who already has code execution on your computer as your user. They can read the data folder directly.
- The fact that your data folder is plain text on disk. It is documented, and you can delete it any time.
- Behaviour of Claude Code, the Claude in Chrome extension, Chrome, or the job sites. Report those upstream.
- Your account being limited by a site because you used automation. See "Responsible use and limits" in the README.
- Findings from automated scanners with no working path through this code.
- The server listens on
127.0.0.1only and answers only aHostof127.0.0.1,localhostor[::1]on its own port, which defeats DNS rebinding. Every request that changes something must come from the app's own origin and carry a customX-CJAheader. Every API call (and the live event stream) must carry the session secret in anHttpOnly; SameSite=Strictcookie. The page contains no secret: the cookie is only set by a one-time launch link (/?launch=<code>) that the launcher opens, each code works once and is replaced, and a second launch of the app reads its next code from.launchin the data folder, which only your user account can read. A bare request for the page, from a website or from another program or user on this computer, gets a page that cannot do anything./api/healthsays only that the app is there. There is no CORS: noAccess-Control-*header is ever sent. - The data folder is created private to your user account, and the profile, answers and run logs are never sent anywhere by the app itself.
- It serves only the files in
public/, and writes only inside the data folder, which by default is outside the app folder (%LOCALAPPDATA%\JobsExpress, or~/.jobs-express). - Agents are started with
claude -p --chrome --setting-sources project --permission-mode dontAskand an allow-list: a named list of Chrome tools (nojavascript_tool, screenshot upload or batch tool),WebFetchfor a few public job-feed domains,WebSearch, andBash(node tools/log.mjs:*). There is noRead,EditorWrite. Your own Claude settings and other MCP connectors are disabled for these runs, and the agent environment is an allow-list that never includes tokens or keys. Claude Code's built-in read-only shell commands (for exampleecho) can still run inside the project folder. The prompt travels on standard input. - The daily schedule registers an operating-system job through fixed scripts and validated fields. The Windows task is created by a constant PowerShell script that receives its values from environment variables. The install button runs only Anthropic's official installer commands, shown to you first.
- Web pages are untrusted input to the agents. The prompts tell them to ignore instructions found in pages, and a prompt cannot be a security boundary on its own, which is why the tool allow-list exists.