What problem does it solve? Adding a new coding agent backend (Codex, Aider, Cursor, etc.) to loop-generator requires implementing the AgentDriver contract correctly — confining edits to the workspace, honoring abort signals, applying feedback, and classifying stop reasons honestly. This Skill guides that work as a feedback loop: scaffold the driver, register it, then iterate on loopgen verify-driver <name> until all four conformance scenarios pass. ## Core Features & Use Cases - Backend classification: Determines whether the target is an agentic CLI, an agentic SDK, or a model server, and routes model servers (LM Studio, Ollama, OpenAI-compatible) to the existing opencode driver instead of writing a new one. - Conformance-driven loop: Maps each failing scenario (reports-name, creates-file, applies-feedback, honors-abort) to a concrete fix, repeating until verify-driver reports PASS. - Copy-paste skeletons: Provides a subprocess CLI skeleton and a model-server skeleton, plus registration, preflight helpers, and test-seam patterns so unit tests never hit the network. - Use Case: A user wants to integrate the Aider CLI as a loop-generator backend. The Skill scaffolds src/drivers/aider.ts from the CLI skeleton, registers it in src/registry.ts, then drives verify-driver aider to green one scenario at a time. ## Quick Start Ask the agent to add a new loop-generator driver for your chosen agentic CLI or SDK and drive it to green with the verify-driver conformance harness.
