# Security

> if you've found a security vulnerability in inference shell or belt.sh, we want to hear about it.

**URL:** https://inference.sh/docs/policy/security-disclosure
**Last updated:** 2026-09-18

---

## reporting a vulnerability

if you've found a security vulnerability in inference shell or belt.sh, we want to hear about it.

**email:** [security@inference.sh](mailto:security@inference.sh)

please include:
- description of the vulnerability
- steps to reproduce
- affected component (api, cli, web, skills, agents)
- impact assessment (what an attacker could do)

we will acknowledge your report within 48 hours and aim to provide a fix timeline within 5 business days.

## reporting abuse

for abuse reports (malicious skills, apps, or agents):

**email:** [safety@inference.sh](mailto:safety@inference.sh)

for dmca takedown requests:

**email:** [dmca@inference.sh](mailto:dmca@inference.sh)

## what we protect

- all skill content is scanned by our security scanner (INF-SEC rules) before serving
- credentials detected in skill content are automatically redacted before serving
- hook tool URLs are validated (https required, no localhost)
- critical security findings block skill publishing automatically
- credential redaction uses hardcoded provider-token patterns (OpenAI, Anthropic, AWS, GitHub, Stripe, and others) — supplemental [gitleaks](https://github.com/gitleaks/gitleaks) rules are currently disabled

## scope

the following are in scope for security reports:

- inference shell api and web application
- belt.sh web application  
- belt cli
- skill store and registry
- agent runtime and execution
- mcp connector proxy
- authentication and authorization

the following are out of scope:

- third-party app provider vulnerabilities (report to the provider directly)
- social engineering attacks
- denial of service attacks
- issues in dependencies already reported upstream

## disclosure policy

- we follow coordinated disclosure — please give us reasonable time to fix before publishing
- we will not pursue legal action against researchers acting in good faith
- we credit researchers in our changelog (unless you prefer anonymity)

## security.txt

our machine-readable security contact is at [/.well-known/security.txt](https://inference.sh/.well-known/security.txt)

## policies

we maintain comprehensive security, incident response, data classification, vendor management, business continuity, and change management policies. full policy documents are available on request — contact [security@inference.sh](mailto:security@inference.sh).