Agent Verify 2.0

Agent Verify 2.0 checks AI agents everywhere they are built and run.

• 124 tested security checks (48 new): MCP tool poisoning, AI agents in CI, payments and deletions without limits, PII sent to models, and checks for 11 agent frameworks
• Pull request checks on GitHub, GitLab and Bitbucket
• VS Code extension and an MCP server for Claude Code, Cursor and Copilot
• Agent Guard runtime rules, prompt-injection tests, compliance evidence packs

Use the Action:
- uses: AI-Blockchain-Ventures/agentverify@v2

Full changelog: https://github.com/AI-Blockchain-Ventures/agentverify/blob/main/CHANGELOG.md

Agent Verify v1.4.0

Agent Verify v1.4.0

Workspaces, role-based access control, signed attestations, and a hardened public/private boundary between the open-source client and the proprietary detection engine.

Workspaces & RBAC

• New organizations/workspaces with four roles — Owner, Admin, Member, Viewer — enforced server-side on every request, not just hidden in the UI.
• Invite members, manage roles, and view a server-written audit log of security-relevant activity (scans, verifications, membership changes, policy application).
• A dedicated in-app Billing tab and expanded Workspaces/Webhooks documentation.

Signed Attestations

• Scan results can now carry a signed attestation (ECDSA P-256) that anyone can independently verify — no shared secret required.

Policy Evaluation

• Built-in policy profiles (standard, high-security, financial-agent, production-infrastructure) can be evaluated against scan evidence from the CLI, GitHub Action, or API.

Webhooks (Configuration Foundation)

• Workspaces can configure webhook endpoints and select which events they care about, with a per-webhook signing secret and an HMAC-SHA256 signature scheme (Agent-Verify-Signature: t=...,v1=...) that receivers will use once automatic delivery is enabled.
• Automatic delivery is not yet enabled — configuring a webhook does not currently send it live events. This is intentionally shipped as a foundation, documented as such, not represented as a completed feature.

Billing

• Dashboard, CLI, and API scans now share one server-enforced monthly quota per account — quota can no longer be bypassed by clearing local browser storage.
• Checkout correctness fix: an already-subscribed account is routed to the billing portal instead of risking a second, concurrent subscription.
• Fixed production workspace-creation and billing-navigation issues so the authenticated dashboard experience matches what's documented.

Reports & Dashboard

• Scan-to-scan comparison, a posture timeline, and refreshed report-view rendering.
• A public, unauthenticated demo scan endpoint and an authenticated fix-verification endpoint — both run entirely server-side.

Security Hardening

• Removed every remaining client-side code path that could pull the proprietary detection engine into a browser-shipped bundle; added a permanent CI check that fails the build if this ever regresses.
• Findings that reference a detected credential now show only a redacted preview — the live secret value is never included in scan evidence or a shared report.

CLI (agentverify@1.4.0)

• Behavior change: --ci mode now uses a distinct exit code 3 for an execution error (bad/missing API key, network failure, missing file), separate from exit code 1 (a real NOT_VERIFIED security finding). Pipelines that only check for a non-zero exit code are unaffected; pipelines branching on the exact code 1 should treat both 1 and 3 as failures, or handle 3 as an infrastructure problem rather than a security finding.
• Packaging fix for the published npm executable.

---

Full Changelog: CHANGELOG.md · CLI Changelog

Install: npm install -g agentverify