Skip to main content

​
1. Agent Review Process

All submitted agents undergo automated & manual analysis checking for code vulnerabilities, prohibited functionality, and adherence to API usage guidelines. Documentation completeness is evaluated against standardized requirements including clear purpose statements, input/output specifications, and limitation disclosures. Agents must demonstrate basic functionality and reliability through preliminary testing before advancing to comprehensive review stages.
Agents are evaluated against domain-specific test suites designed to measure accuracy, response time, and resilience under various load conditions. Stress testing simulates high-volume usage patterns to identify potential failure points and performance degradation scenarios. Comparative analysis positions each agent against existing marketplace offerings with similar functionality to establish relative performance expectations.
Multi-stage safety evaluations probe for harmful outputs, testing agent responses to adversarial inputs and edge cases. Fairness assessments measure performance across demographic groups to identify potential bias in agent responses or recommendations. Independent review panels evaluate high-risk agents with additional scrutiny for applications in sensitive domains like healthcare, finance, and legal services.
Initial screening results are provided within 24 hours of submission, with detailed technical feedback on any failed criteria. Standard review processes for low and medium-risk agents are completed within 5 business days, providing comprehensive evaluation reports. High-risk or complex agents requiring specialized review may take up to 10 business days, with progress updates provided at predetermined milestones.
Developers can submit appeals through a structured process that requires addressing specific rejection reasons with evidence of remediation via emails to [email protected] . Appeals are reviewed by different evaluators than those involved in the initial rejection decision to ensure fresh perspective. Expedited re-review options are available for agents that have addressed all identified issues, with clear guidelines for qualifying for this accelerated path.

​
2. Content Guidelines

The platform explicitly prohibits agents designed for illegal activities, harassment, discrimination, or the generation of misleading content. Restrictions cover agents that could enable unauthorized access to systems, circumvent security measures, or generate spam content. Detailed examples clarify boundaries between acceptable and prohibited use cases, with regular updates reflecting emerging risks and regulatory changes.
All agents must adhere to core principles including transparency in capabilities, respect for user autonomy, and fairness across demographic groups. Output traceability requirements ensure users can understand the basis for agent recommendations and decisions, especially in high-stakes domains. Developers must implement appropriate safeguards proportional to the potential risks associated with their agent’s functionality and intended use cases.
Region-specific requirements address varying data protection regulations including GDPR, CCPA, and emerging privacy frameworks. Industry-specific compliance modules enforce additional safeguards for regulated sectors such as healthcare (HIPAA), finance (PCI-DSS), and education (FERPA, COPPA). The platform provides compliance verification tools that help developers ensure their agents meet all applicable requirements for their target markets.
A tiered rating system classifies agents based on content sensitivity, complexity of outputs, and required user expertise. Professional-only designations restrict certain agents to verified business users in appropriate domains like legal, medical, or financial services. Clear rating display requirements ensure users understand potential content sensitivity before interaction with any agent in the marketplace.
Progressive enforcement mechanisms begin with warnings and remediation periods for minor or first-time violations. Serious violations trigger immediate agent suspension pending investigation and developer response. Repeat violations lead to escalating consequences including extended review periods for future submissions and potential developer account restrictions. All enforcement actions include specific violation details and clear remediation requirements.

​
3. Safety and Security

The platform implements a comprehensive STRIDE threat modeling approach (Spoofing, Tampering, Repudiation, Information disclosure, Denial of service, Elevation of privilege) for all components. Automated static and dynamic code analysis tools scan agent code and configurations during submission and after any updates. Third-party dependency scanning identifies known vulnerabilities in libraries and frameworks used by agents, with automated notifications when critical updates are required.
Reliability is measured through multiple dimensions including uptime percentage, mean time between failures, and response time consistency under varying loads. Chaos engineering practices systematically introduce controlled failures to verify system resilience and recovery mechanisms. Performance degradation tracking identifies subtle issues before they impact user experience, with automated alerts when metrics fall below defined thresholds.
Agents undergo systematic testing against demographically diverse datasets to identify performance variations across different user groups. Standard fairness metrics including equalized odds, demographic parity, and counterfactual fairness are calculated for each agent’s outputs. Regular red-team exercises challenge agents with adversarial inputs designed to elicit potentially biased or harmful responses.
Independent security audits are conducted quarterly by certified third-party specialists examining infrastructure, code, and operational procedures. Continuous automated scanning supplements formal audits with daily vulnerability checks and configuration assessments. Bug bounty programs incentivize responsible disclosure of security issues from external researchers and ethical hackers.
The platform maintains SOC 2 Type II certification with annual audits verifying controls for security, availability, and confidentiality. ISO 27001 certification ensures the information security management system meets international standards for best practices. Additional certifications for specialized industries include HIPAA compliance for healthcare and PCI-DSS for payment processing when applicable.

​
4. Personal Data Handling

Agents should be designed to operate with the minimum data necessary to fulfill their function, with clear documentation of required data fields. Automated data classification identifies and flags potential collection of excessive personal information during the review process. Just-in-time data collection practices ensure information is only gathered when needed rather than preemptively stored.
Multiple anonymization methods including tokenization, hashing, and perturbation are applied based on data sensitivity and use case requirements. Differential privacy techniques add calibrated noise to aggregate data while preserving analytical utility for training and improvement. Pseudonymization systems allow for functional operations while removing direct identifiers from operational datasets.
Granular data retention policies specify maximum storage durations based on data type, purpose, and regulatory requirements. Automated deletion processes permanently remove expired data according to schedule, with cryptographic verification of complete removal. Emergency purge capabilities enable immediate deletion of specific data categories when required for security or compliance reasons.
Regional deployment options ensure data remains within specific geographic boundaries to comply with data sovereignty requirements. Industry-specific configurations implement additional controls for healthcare, financial, and government data with strict localization needs. Transparent documentation specifies data storage locations and transfer mechanisms for all platform services.
Self-service portals allow users to view, export, and manage personal data stored within the platform. Granular permission settings enable users to authorize specific data uses while restricting others. Consent management tracks user preferences with immutable audit logs of all preference changes and data access events.