From harness-claude
Implements continuous identity verification using device trust posture, risk-based session scoring, and identity-aware access proxies for zero trust architecture.
npx claudepluginhub intense-visions/harness-engineering --plugin harness-claudeThis skill uses the workspace's default tool permissions.
> Authentication at login is necessary but insufficient -- continuously evaluate identity
Implements zero trust identity verification using FIDO2/WebAuthn MFA, risk-based conditional access, and CISA-aligned governance. For secure auth in enterprise environments.
Implements continuous identity verification for zero trust using FIDO2/WebAuthn MFA, risk-based conditional access, and CISA Zero Trust-aligned governance. For secure enterprise auth setups.
Guides Zero Trust architecture implementation with ZTNA, micro-segmentation, identity-first security, continuous verification, and BeyondCorp for network security and cloud-native apps.
Share bugs, ideas, or general feedback.
Authentication at login is necessary but insufficient -- continuously evaluate identity confidence using device trust, behavioral signals, and environmental context throughout the session
Traditional authentication verifies identity once at login and trusts the session for its entire lifetime. Session hijacking (stolen cookies, XSS-extracted tokens), credential compromise discovered mid-session, and device compromise after login all exploit this one-time-check model. Google's BeyondCorp (deployed internally after the 2009 Operation Aurora attack by Chinese state hackers) eliminated the trusted network perimeter and replaced it with continuous per-request identity and device verification. The 2022 Uber breach demonstrated that initial authentication (even with MFA) is insufficient when session tokens are subsequently stolen via social engineering -- the attacker MFA-bombed an employee, obtained a session token, and accessed internal systems for hours before detection. The 2023 MGM Resorts breach used social engineering to bypass MFA and then operated freely within sessions that were never re-evaluated.
Evaluate device trust posture before granting access. Collect device signals: OS version (is it patched?), disk encryption status, screen lock enabled, MDM enrollment status, firewall enabled, antivirus status. Define a minimum device posture profile for each access level: accessing public docs requires basic posture; accessing production systems requires fully managed, encrypted devices with current patches. Reject or downgrade access for devices that do not meet the required posture.
Implement risk-based session scoring. Assign a risk score to each session based on signals: IP geolocation change, impossible travel (login from New York and London within 30 minutes), new device, new browser, unusual time of day, high-velocity actions (100 API calls per minute from a normally low-volume user). When the risk score exceeds a threshold, trigger step-up authentication (re-enter password, MFA challenge) or terminate the session. Score continuously, not just at login.
Use identity-aware access proxies. Replace VPN with a reverse proxy that evaluates identity, device posture, and access policy on every request. Google BeyondCorp, Cloudflare Access, Zscaler Private Access, and Palo Alto Prisma Access implement this model. The proxy intercepts every request, validates the user's identity (via session token), checks device posture (via client certificate or agent attestation), and evaluates the access policy before forwarding the request to the backend. No request reaches the backend without passing all checks.
Implement device attestation. Use client certificates (issued by your CA, bound to the device's TPM) or device attestation protocols (Apple DeviceCheck, Android SafetyNet/Play Integrity, Windows Device Health Attestation) to cryptographically verify that the device meets your posture requirements. Self-reported device attributes are insufficient -- they can be spoofed by malware or a rooted device.
Define access tiers by sensitivity. Not all resources need the same level of identity assurance. Tier 1 (public docs): authenticated user, any device. Tier 2 (internal tools): authenticated user, managed device, current patches. Tier 3 (production access, financial data): authenticated user, managed device, MFA within last hour, device attestation, trusted network or VPN. Document these tiers and enforce them consistently across all access paths.
Handle degradation gracefully. When device posture degrades mid-session (e.g., MDM detects the device is rooted or a required certificate expires), the system should reduce access level, not crash. Downgrade from Tier 3 to Tier 1 and notify the user that additional verification is needed to regain full access. Never fail open -- if posture cannot be determined, default to the lowest access tier.
BeyondCorp has four components: device inventory (tracks all devices and their trust level), access control engine (evaluates policies per request), access proxy (intercepts all traffic), and single sign-on (authenticates users). Request flow: user sends request to access proxy, proxy checks SSO for authentication, proxy queries device inventory for device trust level, proxy evaluates access policy (user identity + device trust + resource sensitivity), proxy forwards or denies the request. No VPN, no trusted network, every request is individually authorized. Google published the BeyondCorp papers in 2014-2016, and the model has become the foundation of modern zero trust architecture.
| Signal Category | Signals | Risk Weight |
|---|---|---|
| Authentication age | Time since last MFA, token age | High |
| Location | IP geolocation change, impossible travel, VPN/Tor exit node | High |
| Device | New device, OS change, missing MDM, disabled encryption | Medium-High |
| Behavioral | Unusual API calls, unusual data volume, off-hours access | Medium |
| Session | Concurrent sessions from different locations, rapid IP changes | Medium |
Combine signals into a composite risk score. Weight them based on your threat model. A geolocation change alone might not warrant a challenge, but a geolocation change combined with a new device and off-hours access should trigger immediate step-up authentication.
WebAuthn can be used not just at login but for continuous re-authentication. Silent authentication using a platform authenticator (Touch ID, Windows Hello, Android biometric) provides a low-friction re-verification that confirms the user is still physically present. This is stronger than session timeout-based re-authentication because it proves physical presence, not just session liveness. Use silent WebAuthn challenges before high-sensitivity operations (financial transactions, admin changes, data exports).
A practical session risk scoring system assigns numeric weights to each signal and maintains a running composite score per session. Example: new device (+30), impossible travel (+50), failed MFA attempt (+20), unusual time of day (+10), high API velocity (+15), known Tor exit node IP (+40). Thresholds: below 25 is normal (no action), 25-50 triggers enhanced logging, 50-75 triggers step-up authentication, above 75 terminates the session. Store the score in the session metadata and update it on every request. Decay scores over time (reduce by 5 points per hour of normal activity) to avoid permanent lockout from a single anomalous event.
| Solution | Type | Device Posture | Best For |
|---|---|---|---|
| Cloudflare Access | Cloud proxy | Limited | Simple setup, cloud-only |
| Teleport | Open source | Strong (mTLS) | SSH, Kubernetes, database access |
| Boundary | HashiCorp | Via integrations | Session-based access, credential brokering |
| Pomerium | Open source | mTLS-based | Identity-aware proxy, self-hosted |
| Zscaler ZPA | Commercial | Full MDM | Enterprise, complex compliance requirements |
Authenticate once, trust forever. A session that is valid for 30 days without re-verification gives an attacker 30 days to exploit a stolen session token. Implement session risk scoring and periodic re-authentication, especially for sensitive operations. Long session lifetimes are a gift to attackers.
Self-reported device posture. Trusting the client to report its own OS version, encryption status, or MDM enrollment without cryptographic attestation. An attacker can spoof any self-reported attribute. Require device attestation via TPM-bound certificates or platform attestation APIs.
VPN as the zero trust solution. VPN authenticates at connection time and then grants broad network access to the internal network. This is perimeter security, not zero trust. A compromised device with a VPN connection has the same access as a trusted device. Replace VPN with identity-aware proxies that evaluate policy per-request.
Binary access decisions. Either full access or no access, with no intermediate states. This forces administrators to over-provision access to avoid blocking legitimate users. Implement tiered access that degrades gracefully based on trust level: high trust equals full access, medium trust equals read-only, low trust equals blocked.
No visibility into device fleet. If you do not know how many devices access your systems, their OS versions, or their security posture, you cannot enforce device trust policies. Implement device inventory and posture collection as a prerequisite to continuous verification. You cannot protect what you cannot see.