Do not open a public issue or pull request for a security problem.

Report it privately through one of:

• GitHub private vulnerability reporting — preferred
• Xsolla security: security@xsolla.com

Please include what the issue is, how to reproduce it, and what an attacker could achieve. We'll acknowledge within three business days.

This repository contains agent skills — Markdown instructions that AI coding agents load and act on. It ships no runtime code, so the threat model is unusual and worth stating plainly.

Treat these as security issues:

• Instructions that leak credentials. A skill that tells an agent to put a secret key, API key, or session token somewhere client-side, log it, commit it, or send it to a third party.
• Instructions that weaken verification. Guidance that skips or botches webhook signature verification, disables TLS checks, accepts unauthenticated input as trusted, or recommends a non-constant-time secret comparison.
• Prompt injection in skill content. Text crafted to redirect an agent away from the user's intent — instructions to fetch and execute remote content, exfiltrate files, or ignore prior instructions. This is the main supply-chain risk for this repo: a skill file is an instruction payload.
• Committed secrets. A real key, token, JWT, or private key in the tree or in history. Report it privately, don't push a revert — the credential must be rotated.
• Workflow privilege escalation. A GitHub Actions change that grants a job more token scope than it needs, runs untrusted PR code with write access, or uses an unpinned third-party action.

Not security issues: a wrong API endpoint or field name, a skill that fails to trigger, or documentation that is merely out of date. Those are normal bugs — open an issue.

Skills in this repo reference credentials by environment-variable name only (XSOLLA_PROJECT_API_KEY, XSOLLA_WEBHOOK_SECRET, XSOLLA_PUBLISHER_TOKEN, …).

When contributing:

• Never paste a real key, token, JWT, merchant ID secret, or webhook secret — not in a SKILL.md, a reference, a fixture, or a PR description. CI scans for JWTs, GitHub tokens, sk- keys, and private keys, but it is a backstop, not a guarantee.
• Keep secrets server-side in the guidance you write. Any skill that has an agent put a secret in browser-reachable code is a bug in the skill.
• Sanitize fixtures. Real payloads carry real user emails, IPs, and transaction IDs — replace them with synthetic values, as skills/webhooks-impl/fixtures/ does.

The main branch is the only supported version. Fixes land there and flow to plugin consumers on their next install or update.