magi_skill

Overview

This skill is a deliberation engine that evaluates decisions through three independent perspectives to produce a single, auditable verdict. It runs either internal lenses (Logos/Pathos/Sophia) in Simple Mode or three external engines (Claude/Codex/Gemini) in Engine Mode. It focuses on architecture choices, trade-offs, go/no-go judgments, strategy, and prioritization without writing code. Outputs include a confidence-weighted verdict, a risk register, and next-step handoffs to downstream agents.

How this skill works

Magi frames the decision, has three independent deliberators evaluate the options, collects votes with calibrated confidence scores, then synthesizes consensus or documents dissent. Each perspective provides a one-line rationale and confidence; the system records a weighted verdict, risk register, and an audit trail ID. For high-stakes or ambiguous cases, Magi can escalate to Engine Mode for physically independent model deliberation and prompts human confirmation on split or high-risk outcomes.

When to use it

• Choosing system architecture or major patterns (monolith vs microservices).
• Deciding build vs buy, refactor vs rewrite, or other strategic trade-offs.
• Making go/no-go release or feature approval calls under uncertainty.
• Prioritizing competing requirements or allocating scarce engineering resources.
• Resolving quality attribute trade-offs (performance vs readability, security vs UX).

Best practices

• Provide clear decision scope, constraints, and measurable success criteria before starting.
• Specify urgency and reversibility so Magi can choose Simple or Engine Mode appropriately.
• Request explicit artifact handoffs (e.g., implementer, launch, priority list) to avoid ambiguity after decision.
• Treat split (1-1-1) or unanimous rejection (0-3) as prompts for human intervention.
• Use the risk register and audit trail for compliance and post-mortem reviews.

Example use cases

• Decide whether to adopt a new backend framework given team skills and time-to-market.
• Weigh shipping a feature now versus delaying for additional testing and UX polish.
• Prioritize sprint backlog items when business value and technical risk conflict.
• Resolve a trade-off between encryption overhead and client-side responsiveness.
• Determine go/no-go for beta launch after mixed QA reports.