dgr_skill

Overview

This skill produces audit-ready decision artifacts for LLM outputs as schema-valid JSON. It captures decision context, explicit assumptions and risks, a recommendation with rationale, and a consistency check to support traceability and review.

How this skill works

Given a decision request and an operating mode (dgr_min, dgr_full, dgr_strict), the skill decomposes the problem, surfaces assumptions and risks, generates a recommendation with supporting rationale, and runs a consistency check. The result is a single JSON artifact that meets minimum schema criteria and is suitable for ticketing, incident logs, or audit storage.

When to use it

• When you need an auditable record of why a particular recommendation was made
• For high-stakes or compliance-sensitive decisions that require reviewer-friendly structure
• When teams must surface explicit assumptions and risks before acting
• To standardize decision outputs across models and contributors
• When traceability and machine-validated artifacts are required for later review

Best practices

• Provide clear decision context and any relevant identifiers (ticket ID, policy name) up front
• Choose mode according to risk: dgr_min for speed, dgr_full for balance, dgr_strict for high risk
• Always store the returned JSON artifact in the project’s audit log or ticketing system
• Ask clarifying questions when inputs are incomplete; do not proceed with hidden gaps
• Treat the output as governance support, not a substitute for human judgement or domain expertise

Example use cases

• Producing a review-ready recommendation for a product security triage ticket
• Documenting assumptions and mitigations for a policy change proposal
• Creating an auditable decision record for incident response actions
• Standardizing vendor selection rationale across procurement reviews
• Capturing conservative, review-gated recommendations for high-risk model deployments