v1.18.1

What's Changed

• feat(agent-server): add token-level streaming support to agent server WebSocket by @VascoSch92 in #2751
• fix(skills): Load installed skills in load_user_skills() by @sjathin in #2884
• chore(deps): bump python-multipart from 0.0.22 to 0.0.26 by @dependabot[bot] in #2878
• Add encryption support for mcp_config field in AgentBase by @tofarr in #2900
• Release v1.18.0 by @all-hands-bot in #2908
• fix(tools): Gracefully handle missing PS1 metadata instead of crashing by @VascoSch92 in #2742
• fix(llm): pin prompt_cache_key to conversation id by @VascoSch92 in #2907
• fix(llm): preserve Responses function_call.id for byte-identical repl… by @VascoSch92 in #2916
• fix(conversation): repin prompt_cache_key on switch_profile by @VascoSch92 in #2920
• Add cancel-eval workflow to kill evaluation jobs by @juanmichelini in #2914
• feat: add repository cloning and skill loading methods to OpenHandsCloudWorkspace by @malhotra5 in #2912
• fix(sdk): add security_risk to prompt-based tool calling example by @VascoSch92 in #2746
• feat: Add secrets support for MCP config variable expansion by @jpshackelford in #2873
• Fix CVE-2026-34444: Update lupa to 2.8 by @aivong-openhands in #2925
• fix: pin ACP npm package versions in agent-server Dockerfile by @simonrosenberg in #2927
• fix(security): first-message WebSocket auth to prevent token leakage by @simonrosenberg in #2790

Full Changelog: v1.18.0...v1.18.1