Skip to content

v2.1.0

Choose a tag to compare

View all tags
@kunwarVivek kunwarVivek released this 10 Aug 18:44
· 18 commits to main since this release
v2.1.0
f2cbcda

Added

  • Per-role AI provider configuration — each model role independently configurable with any provider (OpenRouter, Ollama, Together, Groq, Azure, etc.) via openai-compatible + base URL. No hardcoded vendor defaults.
  • MCP sampling as zero-config AI fallback — calling agent IS the model, no API keys needed.
  • validate_work_product — reviewer inspects work products against acceptance criteria (not rubber stamp).
  • decompose_task — PM splits rejected tasks into subtasks (GitHub sub-issues). Reject→decompose→re-assign loop.
  • smart_assign — capability-matched, budget-aware task routing with agentIds[] and roleFilter.
  • converge_project — auto-approve passing, auto-reject failing, auto-decompose fix subtasks. Project-scoped.
  • converge_until_done — multi-iteration convergence with progress reports.
  • get_handoff_context — cross-agent context for subtasks (parent + rejection feedback).
  • cleanup_registry — remove stale agents from the registry.
  • Human-readable work product comments — files, tests, summary on every issue.

Fixed

  • Sprint create requires projectId (no more wrong-project guessing)
  • smart_assign skips stale agents, requires capability match
  • converge_project scoped to project items only
  • AI error messages distinguish key-present-model-missing from no-keys

📦 npm: CI publish pending NPM_TOKEN fix. Install from source or GitHub releases.

Assets 2
Loading