What problem does it solve? Running coding work through external agentic CLIs (Cursor CLI, opencode, Codex CLI, droid) requires consistent rules for auto-approval flags, authentication failures, isolation, timeouts, and output verification, which are easy to get wrong per tool. ## Core Features & Use Cases - Unified invocation contract: Defines seven rules that apply to all four CLIs, including yolo mode, login preconditions, worktree isolation, scoped prompts, timeouts, JSON output, and result verification. - Tool selection guidance: Provides a decision table mapping task needs (Cursor models, multi-provider, OpenAI ecosystem, Factory catalog) to the right CLI and its dedicated skill. - Yolo flag reference: Documents the exact auto-approval flag per tool, such as --force for Cursor CLI and --dangerously-bypass-approvals-and-sandbox for Codex CLI. - Use Case: An orchestrator agent receives a coding task, checks which CLI is installed with command -v, picks the matching tool, and invokes it in an isolated worktree with a timeout and a scoped prompt. ## Quick Start Ask the agent to delegate this coding task to an installed headless coding CLI following the delegate-coding contract, using an isolated worktree, the tool's yolo flag, a timeout, and JSON output capture.
