v0.1.86

🚀 Release Highlights — v0.1.86 (2026-05-13)

🧠 bootstrap_subagent Discriminator Is Now Functional

• orchestrator.coordination.criteria_mode: bootstrap_subagent now runs a between-rounds LLM critic via SubagentManager
• The critic reads the task and each agent's latest answer, emits proposed_criteria as JSON, and the orchestrator merges them into bootstrap_criteria_accumulator.json
• The next round's checklist is augmented automatically, giving the same end state as bootstrap_inline but with criteria sourced from a dedicated critic rather than the answering agents
• The discriminator runs once per unique answer snapshot so unchanged rounds are not re-critiqued

🧹 Session-End Criteria Drain

• Orchestrator._drain_at_session_end forces one final drain before final presentation
• Late stdio JSONL emissions are captured instead of being stranded after the last checklist resolution pass

🛠️ Codex MCP Approval Fix

• codex exec workspaces now get both non-interactive approval bypasses:
  • approval_policy = "never"
  • Per-MCP-server default_tools_approval_mode = "approve"
• This prevents external MCP tools such as submit_checklist, create_task_plan, new_answer, and read_media from failing immediately with "user cancelled MCP tool call"

🧪 Tests

• massgen/tests/test_bootstrap_criteria.py expanded to 35 tests for discriminator behavior and session-end drain
• massgen/tests/test_codex_native_hook_adapter.py::TestCodexWorkspaceApprovalPolicy covers Codex approval config across modes

---

📖 Getting Started

• Quick Start Guide
• Try It:

  pip install massgen==0.1.86
  uv run massgen --config massgen/configs/coordination/bootstrap_subagent_criteria.yaml "Create an SVG of an AI agent coding."

• Inspect emerging criteria at .massgen/massgen_logs/<session>/bootstrap_criteria_accumulator.json

What's Changed

• feat: Part 2 of bootstrap criteria by @ncrispino in #1090
• docs: docs for v0.1.86 by @Henry-811 in #1092
• feat: v0.1.86 by @Henry-811 in #1091

Full Changelog: v0.1.85...v0.1.86