Skip to content

Improvement Backlog

This backlog tracks implementation gaps and quality improvements for the current policy-aware Team · Agent · Workflow composition surface. The design decisions are finalized in the design/2026-06-06-*.md specs; this page tracks the work still outstanding against those decisions.

Current Milestone: Team-First Alpha Cut

Status of major features (0.x alpha):

Feature Status
Team(scope=..., policy=...) Built
team.add_agent(...) Built
run / send().stream() Built
agent.card Built
memory="inprocess\|mem0\|graphiti" Built
workspace= (file tools) Built
gateway= + auth=StaticToken / custom TokenSource MCP path Built
skills=[Skill(...)] structured roster Built
peers= outbound A2A delegation (RemotePeer, OfficialA2ATransport) Built
broad requires_skill routing with required_tags fail-closed disambiguation Built
approval proof bundles and approval_only=True gates Built
coactra.agent.recipes.code_change(...) optional builder Built
ModelResolver / ModelProfile / ModelRoute Built

Outstanding: Team and Agent Layer

  • Memory guardrails: injection cap, deletion/export (GDPR), write policy
  • Workspace run allow-list configuration surface
  • More documented production route recipes for OpenCode/Zen, LiteLLM, and host-managed gateways
  • Documented recipes for OAuth and inbound A2A serving

Outstanding: Workflow Layer

  • richer workflow review and approval UX
  • reusable verification profile libraries on top of the typed verifier/check model
  • planner support for emitting verifier roles and richer workflow candidates
  • broader conformance tests for policy-enforced routing and resume semantics across adapters
  • optional bounded retry/rework policies layered above the single-pass code-change helper