This repository distributes agent skills — instruction files that other people's AI agents execute with their own Apify account and APIFY_TOKEN. A malicious or careless skill can spend someone else's money or leak someone else's data. Every contribution is reviewed with that in mind.

• Skills are executed by other people's agents. A SKILL.md is a set of instructions an AI agent follows with the user's APIFY_TOKEN and permissions.
• Actor runs cost money. Skills trigger paid Apify Actors; a skill can cause unintended spending of platform credits.
• Skills handle scraped data. Results may contain personal or business data; a skill could attempt to exfiltrate them to a third party.

A PR is rejected (and an existing skill removed) if it contains any of the following:

• Sending data to any endpoint, webhook, or inbox outside the user's control.
• Hardcoded tokens, API keys, or other secrets. CI enforces this via scripts/lint_references.py, which fails on any apify_api_ token in a skill's files. A token must never be assembled into a URL either — use the Authorization: Bearer header, which the Apify API accepts everywhere; ?token=… leaks the secret into every access log the request touches. That one no regex can catch, so reviewers read for it (see REVIEWING.md).
• Prompt-injection patterns — e.g. "ignore previous instructions", or instructions that disable approval gates or safety checks of the executing agent.
• Instructions that trigger unbounded spending without explicit user confirmation (cost gate) — see the cost guardrails in skills/_template/references/gotchas.md.
• Asking users to provide credentials in plaintext (chat, files, or command output).

If you find a malicious skill or a security issue in this repository:

1. Open a private security advisory (preferred), or

2. e-mail security@apify.com.

Please do not report security issues via public GitHub issues.