Skill Router Registry is intentionally conservative. It indexes skill metadata for routing; it does not prove a skill is safe.
The supported workflow is:
read SKILL.md metadata -> create compact registry record -> mark trust/risk -> require review before install or execution
The project does not support:
- auto-installing skills from GitHub
- executing discovered scripts
- reading secrets from community skills
- changing auth, shell, or Codex configuration without user approval
- treating registry presence as trust
Open a GitHub issue for:
- a discovered skill being marked too trusted
- missing high-risk indicators
- unsafe install or execution behavior
- misleading docs around trust or review
Do not include secrets, tokens, or private repository contents in public issues.
low: instruction-only skill with no obvious execution or credential risk.review: mentions shell commands, package installs, network access, downloads, auth, or file mutation.high: asks for credentials, tokens, private keys, passwords, or bypassing approvals.
When unsure, prefer the higher risk level.