language-agent is a personal, non-commercial, solo-maintained open-source project — an AI agent that teaches you to read Modern Standard Arabic through a conversation that blends English and Arabic. There is no backend, no accounts, no personal data collected, and no analytics today. This file tells you how to report a security problem and what to honestly expect in return.

Please do not open a public issue or pull request for anything security-relevant. A public report can expose other people before a fix exists.

Report it privately instead:

1. Preferred — GitHub Private Vulnerability Reporting. Go to this repo's Security tab and click Report a vulnerability. That opens a private advisory only the maintainer can see.
2. Fallback — if that button isn't available to you, open a regular issue titled only security: please contact me with no details, and I'll move the conversation somewhere private. Don't put the vulnerability details in that issue.

A useful report usually includes:

• what the problem is and why it's a security concern,
• the affected commit SHA or URL (and the deployed site, if it's live),
• steps to reproduce or a short proof-of-concept,
• the impact you think it has.

This is a hobby project maintained by one person in their spare time. There is no service-level agreement and no guaranteed turnaround. Realistically:

• I'll try to acknowledge a report within about a week, usually sooner.
• I fix what I can, when I can, prioritizing anything that could harm someone running the app.
• I'd rather under-promise than post an enterprise-style SLA I can't keep. If a report goes quiet, a polite nudge is fair.

Coordinated disclosure is a request, not a contract: please give me a reasonable chance to ship a fix before going public, and I'll credit you for the find unless you'd prefer to stay anonymous. I can't offer a bounty or any reward.

In scope: the code in this repository and the behavior of the app it builds — including bugs that could harm a visitor running the deployed site (for example, a way to get malicious code or content to execute in their browser).

Out of scope:

• the hosting provider, CDN, DNS registrar, or other infrastructure once a host is chosen — report those to the relevant vendor;
• vulnerabilities in third-party dependencies themselves — report those upstream (I still want to know if one is actually reachable through this app);
• findings that only work because the source code is public. The source is intentionally open (see below); nothing here relies on the code being secret, so "you can read the source" is not, by itself, a vulnerability.

There is one actively developed line of work: the current main branch and whatever is currently deployed from it. There is no release-version matrix and no back-porting. Until the app is actually deployed somewhere, "supported" effectively means main.

This repo is intended to contain no secrets and no personal data. The sensitivity-levels model (Secret / Private / Security-sensitive / Working / Public) and the rule that an exposed secret gets rotated, not merely deleted (a secret that ever reached a public commit must be treated as burned) are defined once in AGENTS.md under "Disclosure & sensitivity" — that is the single source of truth, and it is not restated here. If you spot a committed credential or personal data, please report it privately as above so the value can be rotated.

Most code here is written by AI coding agents under human review, then squash-merged — that development model is itself part of why the repo is public. The inward-facing guardrails for that work (treat repo / PR / web / dependency text as untrusted data, not instructions; never echo or commit secrets; anti-slopsquatting; never rubber-stamp a review) live in AGENTS.md; Claude Code reads the same content via the @AGENTS.md import in CLAUDE.md. This file is the outward-facing reporting policy; those are inward-facing authoring rules.

The maintainer would prefer this code to be private. It is public only to capture four specific benefits:

• (a) Writing. The maintainer's blog links to real, live code; this repo is a worked example behind that writing.
• (b) A public agentic-dev trail. The commit, PR, and review history is itself a demonstration of how the maintainer builds software with AI agents.
• (c) OSS hosting and tooling. Public repos get free or better CI, static-hosting tiers, dependency and secret scanning, and code-scanning tooling.
• (d) Showcase. It shows the design and engineering craft.

That's the whole list. Being public does not mean this project is audited, regulated, or held to any compliance or external-auditability standard — there is no such requirement and no agreement that imposes one. Auditability here is a courtesy the maintainer keeps for his own sake, and it is deliberately decoupled from the public/private decision.

The practical consequence for security: decisions follow from those four objectives and from a small set of real obligations — don't commit secrets or personal data; don't ship something that harms the people running the app; respond to reports honestly and reasonably promptly. They do not follow from a reflex to lock everything down or make it audit-grade just because the source is visible. Controls are added when they serve a real objective or protect actual users, and kept out when they'd only be enterprise theater for a small client-side app.