codex-deck-terminal

To create an approval-first terminal execution plan, you convert user goals into structured, single-command steps that wait for explicit approval before execution. This enforces safe continuation by proposing exactly one actionable command block per reply.

A controller-reactive step protocol adapts terminal execution plans based on structured feedback like success, failure, or timeouts. It uses stable step_id values to ensure reliable recovery and plan revision after command execution attempts.

You structure terminal commands using specific markdown tags like ai-terminal-plan and ai-terminal-step. This allows a UI controller to render executable step cards, capture outcomes, and parse actionable plan blocks machine-readable format.

Yes, terminal execution plans can be grounded using controller-provided environment context such as cwd, shell, OS release, architecture, and platform. This ensures plans adapt accurately without assuming a specific operating system.

To handle failures and timeouts in a multi-step terminal workflow, the plan uses stable step_id-based recovery to adapt subsequent commands. It processes structured execution feedback to revise the plan safely after a rejection or error.

You need a controller capable of parsing structured tagged markdown and providing environment grounding to use approval-gated command proposals. The controller captures execution outcomes and manages the interactive gating required for safe step continuation.