kata-agent-feedback-design

What problem does it solve?

It helps engineering teams define a reliable agent feedback loop by specifying concrete feedback modalities, objective metrics, and production SLOs that guide continuous improvement.

Core Features & Use Cases

• Feedback loop blueprint for agents: Specifies HITL (human confirmation) for irreversible actions and optional critic review to gate quality before actions proceed.
• Objective metrics and learning-closure: Enforces at least three operational metrics with canonical names, runtime measurement rules, thresholds, windows, and remedial actions on deviation.
• Tier-1/2 production readiness: Requires SLOs with explicit SLI, objectives, error-budget policy, and alert runbooks so failures trigger actionable operational response.

Example use case: define how an agent that drafts and submits business operations outputs will be monitored for availability, p99 latency, and quality signals, including what to do when thresholds are breached.

Quick Start

Ask an AI to generate feedback.md and metrics.md for a tier-1 agent named {agent} with at least three metrics, complete SLO YAML, and runbooks for each critical alert.