Content Policy

The binding rules for what may and may not be published on AI Feeders, and how violations are enforced.

DRAFT — must be reviewed by a qualified lawyer before launch.

Plain-language summary

• Publish original, working AI skills that you have the right to share, with honest descriptions and real previews.
• Never upload malware, hidden instructions that hijack someone's AI assistant, or anyone's secrets or credentials.
• No sexual content of any kind — AI Feeders is a clean platform. No doxxing, no impersonation, no instructions for hurting people.
• Don't re-upload other people's work, flood the site with junk listings, or claim your skill can do things it can't.
• Every upload is scanned automatically before it goes live, and new creators get human review for their first skills.
• Most violations: warning first, then a 7-day suspension, then a ban. Malware and other severe violations: instant permanent ban.
• Every enforcement decision tells you the exact rule and the facts, and every decision can be appealed.

Document status: Draft 1.0 (2026-06-12). The effective date is assigned when this document is published at aifeders.com/content-policy on launch day. Contact: legal@aifeders.com

This policy is part of the Terms of Service and applies to everything you submit: skill packages, prompts, agent configs, workflows, descriptions, images, videos, reviews, replies, profiles, usernames, collections, and reports. The friendly, human-readable companion is the Community Guidelines; if the two differ, this policy controls.

1. What is welcome

• Original AI skills — skill packages, prompts, agent configurations, and workflows that you authored, or have the legal right to distribute under the license you select (MIT, CC-BY-4.0, CC0, or Free-for-personal-use).
• Derivative and remixed work where the source license permits it, with the source credited in the description and a compatible license selected.
• Honest listings — titles, taglines, descriptions, cover images, galleries, and videos that truthfully show what the skill does and produces.
• Skills for every legitimate domain — coding, writing, design, research, productivity, data, education, business, and beyond, for any supported AI platform.
• Clearly labeled security education — defensive security content is welcome subject to the labeling rule in Section 2.4.

You are responsible for everything published under your account, including files inside zip bundles and instructions embedded in markdown.

2. Prohibited content

2.1 Malware and code designed to harm — severe, instant ban

• Any code, script, or instruction with destructive, surveillance, or data-stealing behavior, disclosed or not: droppers (curl piped to a shell, encoded blobs that decode to executables), reverse shells, ransomware, keyloggers, cryptominers, wipers.
• The documented "markdown-as-installer" pattern: setup commands or agent-directed steps placed in SKILL.md or other documentation so that an AI assistant — not the human — executes them.
• Instructions that establish persistence (editing shell rc files, crontabs, agent config files) or that disable security tooling, sandboxes, or our scanning, without the user's clear, informed instruction to do so.

2.2 Prompt-injection payloads — severe, instant ban

Skills must never contain instructions designed to manipulate or harm the downloader's own AI agent. Prohibited techniques include:

• Hidden instructions — directives concealed in zero-width characters, HTML comments, white-on-white or CSS-hidden text, metadata fields, homoglyph swaps, or encoded blobs, intended for the AI to read and the human to miss.
• Exfiltration attacks — instructions telling the assistant to read ~/.ssh, .env files, browser profiles, environment variables, or credential stores and transmit them anywhere; markdown images or links with data-bearing query strings to external hosts.
• Instruction-override attacks — "ignore previous instructions", "do not tell the user", "without asking the user", system-prompt-override framing, or any directive that makes the assistant act against its user's interest or conceal its actions.
• Covert behavior changes — skills whose legitimate function modifies assistant behavior must disclose exactly what they change, prominently in the description. Undisclosed manipulation is treated as an attack.

2.3 Leaked secrets and credentials — blocked at upload; escalates

• No API keys, passwords, tokens, private keys, session cookies, or other credentials in any uploaded file. If they are your own, the secret scanner blocks the upload to protect you — rotate any key that was ever uploaded, because we must assume it is compromised. If they are someone else's, that is a severe violation: instant ban and, where appropriate, a report to the affected provider.

2.4 Instructions for serious harm — severe, instant ban

• No content providing meaningful assistance for serious wrongdoing: building weapons, making drugs or explosives, planning violence, stalking, fraud kits, phishing campaigns, or attacks on people or infrastructure.
• No skills designed to produce illegal content or marketed as tools to bypass AI safety systems for harmful ends.
• Legitimate security research clarification: genuine red-teaming and defensive-security education may be published if it is clearly labeled as security research in the title and description, explains its defensive or educational purpose, and is not packaged as a working attack tool. When in doubt, email legal@aifeders.com before publishing — asking first never counts against you.

2.5 Doxxing and harassment — strike; severe or repeated cases ban

• No publishing or threatening to publish anyone's private information: home address, phone number, government IDs, financial details, non-public email, or private images.
• No harassment, threats, hateful conduct targeting protected characteristics, or encouraging others to do any of these — in listings, reviews, replies, profiles, or bad-faith reports.

2.6 Impersonation of people or brands — strike; deceptive intent escalates to ban

• No accounts, usernames, or listings that misrepresent who you are or your affiliation with a person, brand, or organization — including skill names and cover images styled to look official ("Official Claude Skill", fake vendor logos).
• Parody and fan accounts must be clearly labeled in the display name or bio. Claiming to be AI Feeders staff is a severe violation. Staff badges cannot be self-applied.

2.7 NSFW — not allowed in any form

• No sexual content of any kind. AI Feeders is a clean platform: no pornography, no erotica-generation skills, no sexualized imagery in covers, galleries, avatars, or banners — and no "tasteful" or "artistic" exceptions in the MVP. There is no NSFW toggle and no age-gated section.
• Sexual or sexualized content involving minors, real or generated, results in immediate permanent ban, preservation of evidence, and a report to the relevant authorities (including NCMEC where applicable).
• Graphic gore and shock content are likewise prohibited.

2.8 Plagiarism and re-uploads — strike, plus the copyright process

• No re-uploading skills, prompts, or assets created by someone else and presenting them as yours — whether taken from this marketplace, another marketplace, a paid product, or an open repository whose license you violate.
• No wholesale copying of another creator's listing text, media, or structure.
• Selecting a license you have no right to grant is itself a violation. Rights holders should use the Copyright & DMCA Policy, and copyright strikes follow its repeat-infringer rule.

2.9 Spam and SEO junk — strike; at scale, instant ban

• No keyword-stuffed titles or tag lists, duplicate flooding (the same or trivially varied skill uploaded many times), placeholder or empty packages, listings that exist only to carry outbound links, link farms, or engagement manipulation (fake likes, follows, downloads, review trading, review bombing).

2.10 Misleading capability claims — strike

• Listings must claim only what the skill actually does. Prohibited: invented benchmark numbers, fake endorsements, cover images or demo videos showing output the skill cannot produce, results from a different tool presented as the skill's own, and bait-and-switch descriptions. Illustrative mockups are acceptable only if labeled as illustrations.

2.11 Usernames, profiles, and media — strike; severe content escalates

• Usernames, display names, avatars, banners, and bios are content too: no slurs, no sexual or shock imagery, no impersonation (Section 2.6), no contact-info harvesting ("DM me for the paid version" link farms), and no homoglyph lookalikes of existing creators or staff.
• Review text, replies, collection names, and report descriptions follow the same rules as listings.

2.12 Platform abuse — escalating restrictions

• No circumventing rate limits, upload scanning, or moderation; no automating account creation; no ban evasion through new accounts; no scraping at scale; no interfering with the Service's operation. See also Section 8 of the Terms of Service.

3. Examples at a glance

Illustrative, not exhaustive — the rule text above controls:

4. Quality minimums

Beyond the prohibitions, every published listing must clear this bar — listings that don't are rejected in review or unpublished with a notice, without a strike for a first miss:

1. A working skill. The package must contain real, functional content matching its declared type — a skill package with a valid SKILL.md, a usable prompt, a loadable agent config, or a runnable workflow. Empty shells and broken bundles are rejected.
2. An honest description. The description must explain what the skill does, how to use it, and what AI platform it targets, in the creator's own words.
3. Real preview media. The required cover image and any gallery media must depict the actual skill or its genuine output.

5. How scanning works

Every upload runs through an automated pipeline before anyone can download it: the bundle is validated against a file-type whitelist and zip-safety limits, extracted in a sandbox, scanned for malware with ClamAV, scanned for leaked secrets with rule-based detectors, checked for prompt-injection patterns with the heuristics that flag the techniques in Section 2.2, and its images checked for NSFW content; flagged uploads go to a human moderation queue instead of publishing, new creators' first three skills always receive human review, and published content can be re-scanned and spot-checked as detection rules improve. The technical detail lives in ../05-security/03-upload-content-security.md and ../03-backend/03-upload-pipeline.md; a passing scan earns the listing's "Scanned" badge but is not a safety guarantee — see Section 13 of the Terms of Service.

6. Enforcement ladder

Strikes expire 90 days after they are issued. Copyright strikes are tracked in the same strikes system but follow the Copyright & DMCA Policy repeat-infringer rule: three valid copyright strikes within twelve months terminate the account. Bans remove the account's listings; re-registering to evade a ban is itself a severe violation. The moderation workflow, queue, and audit logging behind this ladder are documented in ../06-admin/02-moderation-system.md.

Every enforcement notice includes: the rule violated, the facts behind the decision, whether automated detection was involved, the current strike count, and how to appeal — the statement-of-reasons commitment in Section 11 of the Terms of Service.

7. Reporting violations

• Every skill, review, and profile carries a Report button, usable by anyone — including visitors without an account. Reports ask for a category and an optional description; reports of illegal content may include your contact details so we can follow up. This is the notice-and-action mechanism under Article 16 of the EU Digital Services Act.
• Copyright owners should use the dedicated process in the Copyright & DMCA Policy.
• Urgent safety issues and anything that doesn't fit the form: legal@aifeders.com.
• Reports are confidential. Retaliating against a reporter violates Section 2.5. Knowingly false reports filed in bad faith may earn the reporter a strike.

8. Appeals

You can appeal any enforcement action within 30 days of the notice:

1. Use the appeal form linked in every enforcement notice (also reachable from the suspension screen), or email appeals@aifeders.com with the case reference.
2. Explain why the decision was wrong or what context was missed; attach evidence if you have it.
3. A human reviewer — different from the original decision-maker wherever staffing permits — re-examines the case and responds with a reasoned decision, normally within 7 days.
4. Upheld appeals reverse the action and remove the strike; the reversal is recorded in the audit log.

9. Changes

We update this policy as new abuse patterns appear in the AI-skills ecosystem. Material changes are announced 30 days in advance per Section 17 of the Terms of Service; changes that close actively exploited loopholes may take effect immediately, with notice.