OpenAgentSkill helps agents and builders discover, compare, audit, and install reusable AI agent skills. The platform is designed to improve decision quality before a skill is used, but it does not make third-party code safe by default.

This policy covers:

• The OpenAgentSkill website and API.
• Registry, ranking, audit, badge, claim, and outcome-feedback surfaces.
• Protected automation routes for indexing, IndexNow, and X posting.
• Public skill metadata generated by OpenAgentSkill.

This policy does not cover:

• Third-party repositories listed on OpenAgentSkill.
• Code, scripts, dependencies, prompts, or install commands maintained outside this repository.
• Claims made by third-party skill authors unless they are explicitly verified.

OpenAgentSkill does not automatically execute third-party skills for users.

Before installing or running a skill, review:

• Source code and scripts.
• Dependency changes.
• Install command.
• Environment-variable usage.
• Network calls.
• Shell execution.
• License and maintenance status.
• Prompt-injection or data-exfiltration risk.

For sensitive work, run new skills in a sandboxed workspace first. Do not expose production secrets, customer data, private keys, or personal data to unreviewed skills.

Trust Score and audit pages are decision-support signals. They combine metadata such as GitHub activity, README/SKILL.md completeness, license clarity, install safety, permission surface, and real agent outcome feedback.

They are not:

• A formal security certification.
• A guarantee that a repository is safe.
• A replacement for manual review.
• A promise that install commands will work in every environment.

Please open a private security report on GitHub if available, or contact the maintainer with:

• Affected URL, endpoint, or repository.
• Steps to reproduce.
• Impact and severity.
• Relevant request/response samples.
• Whether any secret, token, private data, or user account was exposed.

Do not publicly disclose exploitable issues until the maintainer has had a reasonable chance to investigate.

If a listed skill appears unsafe, open an issue with:

• Skill URL on OpenAgentSkill.
• Source repository URL.
• Why it is risky.
• Evidence, such as unsafe shell commands, token access, network exfiltration, malicious dependencies, or misleading metadata.

OpenAgentSkill may mark a skill as blocked, experimental, or needs review while the report is investigated.

Privileged routes must stay server-side and protected by bearer secrets or OAuth authorization. Never expose:

• Supabase service-role or secret keys.
• X OAuth client secrets or refresh tokens.
• GitHub tokens.
• Indexer or cron secrets.

Public browser code should only use public Supabase anon credentials protected by Row Level Security.