v1.9

agent0ai/agent-zerov1.9Apr 13, 2026by github-actions[bot]

AI Summary

Addresses critical security vulnerabilities regarding SSRF and path traversal, introduces a new A0 CLI Connector plugin, and restores lexical skill matching functionality.

Key Highlights

  • Fix SSRF in `document_query` remote fetching (CVE-2026-4308)
  • Block path traversal in `download_work_dir_file` (CVE-2026-4307)
  • New A0 CLI Connector plugin for host connectivity
  • Restore lexical trigger-based skill matching

New Features

  • `a0-setup-cli` built-in skill for connector setup
  • Browser Agent model preset selection
  • Redesigned messaging integration settings UI

Full Release Notes

# Agent Zero Release Notes

## Security Fixes

- **Fix SSRF in `document_query` remote fetching (CVE-2026-4308)** — Remote document fetching now validates URLs before any network request, blocking localhost and non-public IP targets, validating redirect hops, disabling implicit proxy trust, and enforcing a strict size cap. Third-party loaders no longer receive attacker-controlled URLs directly; content is prefetched and parsed from trusted local bytes. A follow-up fix restores compatibility with public sites that rejected the changed request fingerprint.
- **Block path traversal in `download_work_dir_file` (CVE-2026-4307)** — Download requests whose resolved path escapes the runtime base directory are now rejected before any file access, preventing arbitrary file reads.

## New Features

- **A0 CLI Connector plugin** — New built-in plugin lets the host-side A0 CLI connect to Agent Zero over authenticated HTTP and WebSocket, with capability discovery, chat/context lifecycle endpoints, log streaming, and remote editing, code execution, and file-tree bridging.
- **`a0-setup-cli` built-in skill** — Guides users through host-side A0 connector setup with installer-first guidance, container guardrails, and fallback install paths. Updated with Flare Tunnel connection guidance.
- **Restore lexical trigger-based skill matching** — Lightweight trigger-word scoring is back in `search_skills()`, re-enabling `skills_tool:search` and lexical relevant-skill recall for the current user message without requiring vector-DB skill recall.
- **Native chat controls for messaging integrations** — Telegram, WhatsApp, and email threads now support shared transport-level commands (`/project`, `/config`, `/send`, `/queue send`) for managing the active chat directly from within each integration.
- **Browser Agent model preset selection** — The Browser Agent can now use a dedicated `_model_config` preset for browser runs instead of always using the main model, a highly requested feature.

## UI & UX Improvements

- **Redesigned messaging integration settings** — Email, Telegram, and WhatsApp settings panels have been rebuilt with clearer step-based setup flows, guided first-run experiences, provider presets for email, safer access warnings, richer test feedback, and responsive layouts. Advanced email options (server, routing, scheduling) are moved behind an Advanced section.
- **Componentized model config** — The model configuration UI has been refactored into components with the store split into mixins and unified API key management.

## Other Improvements

- Updated plugin skill guidance to formalize `install()`, `uninstall()`, and `preupdate()` requirements when dependencies are involved.
- Added contributor sharing and fork safety documentation.