FAQ
Frequently asked questions about Oathe — how it works, what it costs, and common integration questions.
General
Is Oathe free?
Yes. Oathe is free to use. There is no API key, no sign-up, and no usage limit for public repositories. Submit a skill URL and get a trust score.
What does Oathe actually do?
Oathe installs your AI agent skill (MCP server, plugin, or tool) in an isolated environment, monitors its runtime behavior, applies threat detection patterns, and uses AI behavioral analysis to produce a trust score from 0 to 100 with a corresponding verdict.
What is a “skill” in Oathe’s context?
A skill is any executable component that an AI agent can invoke. This includes MCP servers, LLM tool plugins, function-calling endpoints, and similar constructs. If an AI agent can call it, Oathe can audit it.
Auditing
How long does an audit take?
Most audits complete in 30 to 90 seconds. Complex skills with large dependency trees may take up to 2 minutes. You can track progress by polling the audit status endpoint or connecting via WebSocket.
What URLs are supported?
Oathe accepts:
- GitHub repositories:
https://github.com/org/repo - ClawHub registries:
https://clawhub.com/org/skill - Monorepo subdirectories:
https://github.com/org/repo/tree/main/packages/my-skill
The URL must point to a publicly accessible repository or registry entry.
Can I audit private repositories?
Not yet. Oathe currently supports public repositories only. Private repository support is on the roadmap.
Can I audit a local skill that is not on GitHub?
Not directly. Oathe requires a publicly accessible URL to clone and install the skill. For local development, push your code to a public repo (even a temporary one) and audit it from there.
What happens if I submit the same URL twice?
Oathe deduplicates audits. If the skill at that URL has already been audited at the same commit, the existing report is returned with "deduplicated": true. To force a new audit, set "force_rescan": true in your POST request.
Scores and Verdicts
What is the trust score?
A number from 0 to 100 representing the overall security posture of the skill. It is calculated as a weighted sum of six dimension scores: prompt injection, data exfiltration, code execution, clone behavior, canary integrity, and behavioral reasoning. See Scoring Dimensions for details.
What is the difference between verdict and recommendation?
- Verdict describes the risk level: SAFE, CAUTION, DANGEROUS, or MALICIOUS. It is a classification of the skill’s behavior.
- Recommendation describes the suggested action: INSTALL, INSTALL_WITH_CAUTION, REVIEW_BEFORE_INSTALL, or DO_NOT_INSTALL. It is guidance for the consumer of the skill.
Both are derived from the trust score but serve different purposes — verdict is for categorization, recommendation is for decision-making.
What score thresholds map to each verdict?
| Verdict | Score Range |
|---|---|
| SAFE | 80 - 100 |
| CAUTION | 50 - 79 |
| DANGEROUS | 20 - 49 |
| MALICIOUS | 0 - 19 |
Integration
How often should I re-audit?
Re-audit on each release of your skill. If you use GitHub webhooks (see CI/CD Integration), this happens automatically on every push or release event.
Do I need an API key?
No. The Oathe API is open and does not require authentication for public audits.
Can I embed the results in my README?
Yes. Use the Oathe trust badge. See Trust Badge for Markdown and HTML syntax.
Does Oathe execute my skill’s code?
Yes. Oathe installs and runs your skill in an isolated environment to observe its actual runtime behavior. This is what makes it a behavioral scanner rather than a static analyzer. The environment is destroyed after the audit completes.