ruvnet / ruvnet/ruflo

Integrate MidStream reflex handoff with deliberative agent loop

Open
#3,246 0 comments 0 reactions 0 assignees View on GitHub
Dominant language
TypeScript
Stars
72.7k
Forks
8.6k
Avg merge
2d 23h
Merged PRs (30d)
83

Description

## Goal
Integrate MidStream PR 107's `ReasonerHandoff` into Ruflo as an evidence-only realtime control input.

## Contract
Ruflo consumes ordered reflex events and cancellation accounting from MidStream, then decides whether to continue, pause, replan, or discard pending reasoning work. MidStream reflex receipts must never grant tool or execution authority.

## Required cases

1. user interruption during long reasoning
2. repeated interruption burst
3. backchannel while reasoning continues
4. cancellation before tool execution
5. cancellation after a tool proposal but before RVM authorization
6. stale and reordered events
7. queue overflow
8. multi-agent delegation while an interrupt arrives

## Metrics
Report acknowledgement latency, time to useful feedback, task quality, cancelled token cost, wasted frontier calls, tool correctness, resume latency, stale event handling, and total model cost.

## Acceptance
At least 40% lower interruption acknowledgement latency than the monolithic baseline, task quality within 2 absolute points, cancellation waste below 10% of total deliberative inference, zero unauthorized tool calls, and no privilege derived from the MidStream receipt.

## Governance
MetaHarness issue 301 Track D owns matched evaluation. RVM remains the privileged execution boundary. No autonomous merge or deployment.

Contributor guide

Open the contributing guide

Research direction

Start by reviewing MidStream PR 107 and its ReasonerHandoff contract, then trace Ruflo’s deliberative loop and the RVM authorization boundary. Use MetaHarness issue 301 Track D for the matched evaluation; done means the required interruption cases pass the stated latency, quality, cancellation-waste, tool-safety, privilege, and cost criteria.

Written by the indexing model from the issue text.

Assessment

Tech stack
typescript
Domain
ai, distributed-systems
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Active
Clarity
Mostly clear
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.