v2.24.1 — fleet-style orchestration and agent-quality upgrades

Highlights

• Added dispatch_plan_async for fleet-like dependency-aware async orchestration, so the orchestrator can register a multi-item work graph and automatically unblock ready tasks.
• Upgraded precipitated T2 role generation with
  ole-template-quality, producing richer agent definitions with workflow, collaboration, and output sections instead of thin placeholders.
• Added prompt-budget utilities plus related MCP, council-runner, and test coverage improvements.
• Fixed the Windows async runtime proxy timeout by using the heartbeat-based budget rather than the shorter activity timeout.

Included commits

• feat: add async-plan-dispatch, role-template-quality, prompt budget, and mcp-server improvements
• fix: increase proxy timeout to heartbeatTimeoutMs to fix Windows async spawn P1 (#568)

v2.24.0

Optional models, async worker reliability, T2 auto-repair. See CHANGELOG.md for details.

v2.23.4 — Azure CLI path fallback for ADO VCS

Bug Fixes

• Azure CLI path fallback: ADO auth now resolves az via explicit executable lookup and Windows Azure CLI install paths, so VCS tools still work when the MCP process starts with an incomplete PATH.
• Actionable Azure CLI diagnostics: when Azure CLI still cannot be found, errors now report executable resolution diagnostics and checked fallback locations instead of only saying 'az' is not recognized.
• Existing VCS diagnostics and timeout protections from v2.23.3 remain in place.

v2.23.3 — VCS tool reliability and diagnostics

Bug Fixes

• Azure DevOps timeout resilience: added explicit timeouts for az account token acquisition and ADO HTTP requests so degraded ADO or Azure CLI sessions no longer hang the whole VCS tool path.
• VCS config diagnostics: added vcs_config_diagnostics plus richer VCS failure output showing the resolved vcs.json, configured provider, git remote, and provider-cache state.
• Provider cache correctness: VCS provider caching now keys on workspace, resolved config path, and git remote to avoid stale provider reuse after repo/config changes.

v2.23.2 — Windows ACP zombie process fix

Bug Fixes

• Windows ACP process cleanup: use taskkill /T /F for shell-spawned ACP processes so Copilot/Claude child processes do not survive wrapper-shell exit.
• Runtime shutdown hygiene: drain AcpProcessPool on SIGTERM/SIGINT and clean up the previous runtime PID tree at startup.
• Pool exit cleanup: shut down all adapters on host process exit to prevent stale warm-pool workers accumulating across sessions.

v2.22.1 — Cold Start Fix

Fix concurrent cold start race — pool reuses initializing adapters instead of replacing them. Init timeout 15s->30s. npx github:cloga/optimus-code#v2.22.1 upgrade

v2.22.0 — User-Level Resource Sharing

Shared resources (roles, skills, config, dist) now fall back to ~/.optimus/ across all projects. Upgrade all registered projects in one command. npx github:cloga/optimus-code#v2.22.0 upgrade

v2.21.1 — Upgrade All Projects

upgrade now defaults to upgrading ALL registered projects. Use --single for current-dir only. npx github:cloga/optimus-code#v2.21.1 upgrade

v2.21.0 — User-Level Runtime + Stability

User-level HTTP Runtime (cross-project), crash guard, session concurrency limit, MCP auto-reconnect. npx github:cloga/optimus-code#v2.21.0 upgrade

v2.20.2 — Runtime Stability

Stability

• Crash guard: HTTP runtime survives uncaught exceptions
• Session concurrency limit: max 10 sessions per ACP process
• MCP proxy auto-reconnect on runtime server crash

Bug Fixes

• Fixed auto-start TDZ bug preventing runtime server launch
• Throw actionable error instead of silent fallback on runtime failure

Upgrade: npx github:cloga/optimus-code#v2.20.2 upgrade