Permission review
Every skill declares required permissions. High-risk permissions (filesystem, network, secrets) trigger mandatory human review before public listing.
Report security issues What not to include Do not put OAuth secrets
Security
Security-first by design.
SkillHub treats security as a core operating principle, not an afterthought. Every skill is reviewed, every runtime call is governed, and every secret stays protected.
Principles
How we protect the platform
Runtime governance
Project-scoped API keys, per-call audit trails, rate limiting, and budget enforcement. No anonymous invocations reach production runtimes.
Secret isolation
OAuth secrets, API tokens, and private keys never appear in public manifests, logs, or error responses. Runtime secrets are injected at execution, not stored in contracts.
Incident response
Critical vulnerabilities trigger immediate suspension. Affected developers receive automated notifications with mitigation guidance.
Responsible disclosure
Report security issues without exposing secrets.
Use the public support path to request a secure disclosure channel during Launch Preview. Do not put sensitive data into public reports.
What to include
- ✓Affected public URL or skill slug
- ✓Impact summary and severity estimate
- ✓Reproduction steps (without real secrets)
- ✓Your preferred secure contact method
What not to include
- ✗OAuth secrets or API tokens
- ✗Passwords or private keys
- ✗Customer data or PII
- ✗Exploit payloads or proof-of-concept code
security-report.md
## Security Report
affected: /skills/browser-research
severity: medium
type: permission-escalation
description:
Skill declares read-only in manifest,
but runtime attempts filesystem writes.
contact: researcher@example.com
---
# No real tokens or secrets included
Response timeline
What happens after you report
01
Acknowledgment
We confirm receipt within 24 hours and provide a secure channel if needed.
< 24h02
Triage
The security team assesses severity, scope, and affected systems.
24–72h03
Mitigation
Critical issues receive immediate action. Non-critical issues enter the sprint queue.
1–7 days04
Resolution
Fix deployed, affected parties notified, and public advisory published if appropriate.
VariesCurrent posture
Platform security status
Active
Skill review
All submitted skills undergo automated + human review
Active
Runtime isolation
Per-project key scoping, rate limits, and audit logging
Active
Secret handling
Zero secrets in public surfaces, encrypted at rest
Active
Incident pipeline
Monitoring, alerting, and automated suspension triggers
Need help with something else?
Explore the Skill API or read Docs & Guides to learn more about the platform.