Security Policy
Relevant source files
The following files were used as context for generating this wiki page:
Opificia AI implements a security framework focused on vulnerability reporting, data sovereignty, and safe AI agent operations. The policy defines how the project handles security disclosures and establishes design commitments to protect user data and infrastructure. Because Opificia operates AI agents that read tickets and run code, the project maintains a broad attack surface that includes sandbox isolation and prompt injection defenses.
Sources: SECURITY.md:1-5, SECURITY.md:65-68, README.md:11-15
Vulnerability Reporting
The project requires private disclosure for all security vulnerabilities. You must not open public issues, pull requests, or Discussions for security flaws.
Reporting Channels
You can report vulnerabilities through two primary channels:
- GitHub Private Reporting: Use the "Report a vulnerability" feature in the Security tab of the repository.
- Email: Send reports to security@opificia.ai.
Sources: SECURITY.md:10-21
Reporting Requirements
A comprehensive report includes the following details:
- The affected component, file, or commit.
- An explanation of what an attacker can do and the necessary conditions.
- Minimal steps to reproduce or a proof of concept.
- Credit preferences for the eventual advisory.
Sources: SECURITY.md:23-31
Response Timeline
The project adheres to the following response targets for reported issues:
| Step | Target |
|---|---|
| Acknowledgement | Within 5 working days |
| Initial Assessment | Within 10 working days |
| Coordinated Disclosure | 90 days (default) |
Sources: SECURITY.md:35-42
Security Architecture and Scope
The security model covers the interaction between AI agents, sandboxed environments, and external data sources.
In-Scope Security Areas
The project prioritizes the following technical areas for security research:
- Sandbox Escapes: Preventing agents from accessing the host, network, or other tenants beyond allowed boundaries.
- Prompt Injection: Defending against malicious content in tickets or repositories that triggers unauthorized code execution or data leaks.
- Credential Exposure: Ensuring secrets, tokens, and keys do not leak into logs, prompts, or pull requests.
- Supply Chain: Monitoring for compromised dependencies or build tampering.
- Human Approval Bypass: Preventing any automated merge or deployment that skips configured human review.
Sources: SECURITY.md:49-62, SOVEREIGNTY.md:37-40
AI Agent Security Rules
Agents must follow strict operational constraints to prevent unauthorized changes to the system core.
Agents are explicitly forbidden from modifying CI workflows, CODEOWNERS, MAINTAINERS.md, AGENTS.md, or security configurations.
Sources: plugins/harness-claude/src/prompts.ts:46-52, AGENTS.md:24-26
Data Sovereignty and Privacy
The security policy is reinforced by "Sovereignty Commitments" that ensure the operator maintains control over data and infrastructure.
Design Commitments
- No Phone-Home: Telemetry is disabled by default. If enabled, it only connects to operator-configured endpoints.
- Local Data Storage: Prompts, code, and logs remain in destinations defined by the operator.
- Air-Gap Compatibility: The software works without outbound internet access.
- Human Control: No agent can merge code without recorded human approval.
Sources: SOVEREIGNTY.md:19-24, SOVEREIGNTY.md:33-40, README.md:21-25
Safe Harbor and Research
The project provides a safe harbor for security research conducted in good faith. Research is protected if it:
- Avoids data destruction or service degradation.
- Uses only the researcher's own accounts or test data.
- Stops once a vulnerability is demonstrated.
- Allows the project reasonable time to fix the issue before public disclosure.
Sources: SECURITY.md:72-83
Summary
The Opificia AI security policy integrates traditional vulnerability handling with modern agentic AI safety. By enforcing human-in-the-loop approvals and strict sandbox isolation, the project mitigates the risks inherent in automated software factory operations. All security efforts support the primary goal of providing a self-hostable, sovereign automation platform.
Sources: SECURITY.md:65-70, README.md:11-15