v2.1.4
Highlights
- Org Runtime v2:
monomind org run— a daemon-controlled multi-agent org runtime replacing the deprecated prompt-orchestratedrunorg. Adds intra/inter-org routing, a real interval scheduler, policy-gated tool/file/web/budget enforcement with audit trail, and per-agent provider env (subscription, api-key, Bedrock, Vertex). - Human-in-the-loop: agents can now
ask_humanmid-run — questions persist, appear in a global Human Input inbox, and answers deliver live (or queue offline) back to the org. - Live org dashboard: WebSocket-driven view of agents, chats, tool calls, assets, and inter-org messages, with an animated org-chart activity layer and a Files tab (version history + diff view).
- Dashboard hardening: closed ~20 XSS gaps from unescaped
onclickinterpolation, added auth-token attachment to client write requests, fixed a leak of ~900 orphan dashboard servers oncontrol start, and removed the standalone MonoAgent tracking tab (monoagent now has its own app). - Full audit backlog closed: all P0–P3 findings (116 total) from a full-repo review resolved, spanning monograph, hooks, orgrt, and CLI.
Fixes
monobrowse: eval batch commands no longer lose their own quoteshooks:deriveSubagentSuccessno longer a false-blind text-keyword checkmastermind:createorg/runorg/org-settingsmigrated off the deprecated v1 org shapebrowse: screenshot no longer broken by global--formatflag leak- Native crash at CLI exit fixed (stopped starting WorkerManager on every invocation)
knowledge_ingestMCP tool: path input now validated against traversal/escape
Full changelog: v2.0.0...v2.1.4
Packages
monomind@2.1.4(npm)@monoes/monomindcli@2.1.4(npm)