n-agentic-harnesses-codex

What problem does it solve?

This Skill helps teams and developers design, evaluate, and improve agentic harnesses so systems avoid brittle tool-calling, weak permission models, unresolved workflow state, and missing evaluation plans that lead to unsafe or unmaintainable agent behavior.

Core Features & Use Cases

• Architecture & Shape Selection: Recommend single-agent vs multi-agent patterns, subsystem boundaries, and transport/deployment considerations.
• Tooling, Permissions & State: Define tool registries, approval gates, permission tiers, session durability, retry/idempotency policies, and state machines.
• Evaluation & Observability: Produce evaluation plans, replayable evals, acceptance criteria, and observability/UX guidance to detect failures and regressions.
• Use Cases: Designing a copilot with tool access, auditing an existing agent runtime for gaps, or producing a phased MVP and verification plan for a workflow orchestrator.

Quick Start

Use the n-agentic-harnesses-codex skill to design a lean, testable agentic harness for a tool-calling assistant that includes a tool registry, permission and approval policies, durable sessions with retry semantics, and an evaluation plan.