mcpField
R
ralplan

ralplan

github.com/zereightAgent workflows2.1k10 installsUpdated today
Repository

About ralplan

ralplan is a consensus-planning skill that coordinates Planner, Architect, and Critic agents until a plan is approved. It is for scoped implementation work that benefits from explicit alternatives, review, and testable criteria before execution.

Key features

  1. 01Planner, Architect, and Critic review
  2. 02Options with pros and cons
  3. 03Iterative approval loop
  4. 04ADR in the final plan
  5. 05Deliberate mode with pre-mortem

Use cases

  1. 01Scoping a vague implementation request
  2. 02Comparing architectural alternatives
  3. 03Preparing an agreed plan before execution

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/zereight/gitlab-mcp --skill ralplan

After installing, type/ralplanin Claude Code to confirm it loaded.

Details

From the project README

Ralplan (Consensus Planning)

Shorthand for /plan --consensus. Triggers iterative planning with Planner, Architect, and Critic agents until consensus is reached.

Flags

  • --deliberate: Forces deliberate mode for high-risk work. Adds pre-mortem (3 scenarios) and expanded test planning.

Workflow

  1. @planner creates initial plan with RALPLAN-DR summary:
    • Principles (3-5)
    • Decision Drivers (top 3)
    • Viable Options (>=2) with pros/cons
  2. @architect reviews for architectural soundness
  3. @critic validates quality and testability
  4. Loop until critic approves (max 5 iterations)
  5. Final plan includes ADR (Decision, Drivers, Alternatives, Why chosen, Consequences)

Pre-Execution Gate

Vague execution requests (e.g., "ralph improve the app") are redirected through ralplan first to ensure explicit scope, testable criteria, and multi-agent consensus.

Passes gate (specific enough): prompts with file paths, function names, issue numbers, numbered steps, or acceptance criteria.

Gated (needs scoping): prompts with only vague descriptions and no concrete anchors.

After Approval

  • Execute via /team (parallel agents, recommended) or /ralph (sequential with verification)

Comments

Markdown supported