Content Policy

Last updated: 2026-04-26 · Effective: 2026-04-26

This Content Policy describes what you may and may not publish to the Amitte registry. The submission pipeline enforces some of these rules automatically; the rest are enforced via the Takedown Policy and account suspension under the Terms of Service.

We try to be specific about what's prohibited so you can publish with confidence. When in doubt, file an issue tagged policy-question and ask before publishing.

What you may publish

The registry exists for content that helps AI agents do useful things. Welcome categories include:

• Step-by-step task guides (deploy an app, ship a release, audit a codebase).
• Domain expertise references (legal templates, compliance checklists, SRE runbooks).
• Tool / API authoring patterns (writing MCP servers, designing tool contracts).
• Prompt engineering techniques.
• Code review checklists, security audit guides.
• Architectural patterns and post-mortems.

Skills should be specific, actionable, and honest about limits. Generic surveys of large topics tend to score poorly with the judge LLM.

What is prohibited

The following content is not allowed on the registry. Publishing it may result in immediate removal and account suspension.

1. Sexual content involving minors

Zero tolerance. Any content depicting, describing, or facilitating sexual activity involving minors (under 18) will be removed immediately. We report such content to the National Center for Missing & Exploited Children (NCMEC) and to local authorities as required by law.

2. Malware, exploit kits, ransomware

You may not publish:

• Working ransomware, wipers, rootkits, or other destructive payloads.
• Step-by-step guides to deploy or operate malware.
• Exploit kits packaged as skills.
• Instructions to bypass security controls on systems you don't own.

You may publish:

• Defensive security knowledge (incident response, hardening, threat detection).
• Educational content about how malware classes work, presented neutrally.
• Pen-testing methodologies aimed at authorized engagements.

The line: defensive understanding is fine; offensive recipes targeting systems you don't own are not.

3. Credential theft and exfiltration

Skills that teach how to steal credentials, exfiltrate data, or compromise accounts of people who haven't authorized it are prohibited. This includes:

• Phishing kits and templates.
• Credit card theft / "carding" guides.
• Account takeover playbooks.
• "How to read a target's private files" guides.

Defensive content (how to detect phishing, how to harden against ATO) is welcome.

4. Prompt injection payloads

Direct prompt-injection payloads (artifacts that attempt to override an agent's instructions, exfiltrate context, or manipulate downstream behavior) are blocked at the submission pipeline's prompt-injection-lint stage. Patterns we reject include:

• "Ignore previous instructions" / "disregard prior context".
• Requests to reveal system prompts.
• Hidden Unicode bidirectional control characters.
• Embedded "role": "system" payloads in markdown.

Educational content about prompt injection (defenses, taxonomies, examples for security training) is allowed if presented as analysis, not as a working payload.

5. Copyright and IP violations

Don't publish content you don't have the right to license under the SPDX identifier in your manifest. This includes:

• Wholesale copies of paid courses, books, or proprietary documentation.
• Code lifted from non-permissive sources without attribution.
• Trademarked logos or branding without permission.

Quoting short excerpts for criticism, analysis, or fair use is fine. When in doubt, link to the source rather than copying.

6. Personal data without consent

Don't publish skills that include personal data (PII) about identifiable individuals who haven't consented. Don't publish content that helps doxx someone, harass someone, or facilitate stalking.

7. Hate speech, harassment, threats

Content that promotes violence against, dehumanizes, or threatens individuals or groups based on protected characteristics is prohibited.

8. Misinformation in safety-critical domains

Skills that present demonstrably false information as guidance in domains where errors can hurt people — medicine, mental health, legal advice for serious matters, financial advice for vulnerable populations — are prohibited if the inaccuracies are material.

We're not the arbiter of all truth. We are the arbiter of "is this skill going to get someone killed, sued, or financially ruined."

9. Spam and low-quality content

The judge LLM scores content quality. Skills that score below threshold are flagged in /policy/takedown review. Patterns we treat as spam:

• Auto-generated SEO chum.
• Content padded with filler to reach length minimums.
• Duplicate / near-duplicate publishes under variant names.
• Skills whose only purpose is to advertise a paid product.

You may reference paid products neutrally if they're load-bearing for the skill's instructions; you may not publish a skill that's just a sales pitch.

How we enforce

Mechanism	What it catches
Submission pipeline	Schema, prompt-injection lint, license blocklist, coarse policy patterns (CSAM, malware-ops, exfil, CC-theft heuristics)
LLM judge	Adversarial score (safety) + content score (quality). Below thresholds → reject or flag.
User reports	The "Report" button on every skill detail page → triaged by maintainers.
Takedown requests	See Takedown Policy.
Routine review	Periodic re-evaluation as judge models improve.

Appeals

If your content is removed and you believe the removal was in error:

1. Email policy@amitte.com within 14 days of removal.
2. Include the skill ID, the version, and a written explanation of why you believe the removal was incorrect.
3. We will review and respond within 14 days.

Repeat appeals on substantively similar content will not be re-reviewed.

Changes

This policy will evolve as the registry grows and we encounter new cases. The "Last updated" date reflects the most recent change. Material changes are announced in the Changelog.

Contact

policy@amitte.com for content questions, or open an issue at github.com/amitte-ai/amitte/issues tagged policy-question.