Skip to content

Multi-agent review pipeline after task completion #16

Closed

Description

@dominicnunez

Summary

After Ralph completes PRD tasks, run a multi-agent review pipeline before declaring done.

Inspired by ralphex's 4-phase approach but simplified:

Phase 1: Ralph executes tasks (current behavior)
Phase 2: Review agents run in parallel (new)
Phase 3: Fix confirmed issues (new)  

Review Agents

Run N review agents concurrently, each with a focused prompt:

  • quality — bugs, security, race conditions
  • testing — test coverage gaps, weak assertions
  • simplification — over-engineering, dead code

Each agent reviews the diff (not full codebase), reports findings. Then a fix pass addresses confirmed issues.

Why

This bridges Ralph and Willie into one flow. Instead of "run Ralph, then manually run Willie," the review is built into the build pipeline. Ralph builds → agents review → fixes applied → done.

Config

[ralph]
review = true              # enable post-task review
review-agents = quality,testing,simplification

Or sfk run --review to enable for a single run.

Relation to Willie

Willie remains the standalone deep audit tool. This is a lightweight review pass — faster, scoped to changes only, fewer agents. Willie is the full codebase scan.

Activity

  1. dominicnunez commented on Feb 21, 2026

    @dominicnunez
    OwnerAuthor

    Closed as duplicate of #24 - RFC: Springfield Ultra

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions