spec-kitty-runtime-next

Mission orchestration for stalled workflows is automated by running the canonical runtime control loop, which applies doctrine-aware action context loading to interpret decision kinds, resolve guard failures, and advance blocked steps. It repeatedly executes the returned prompt to drive the mission forward.

A runtime control loop drives mission execution by loading an agent profile at initialization and refreshing governance context at each step boundary. It uses bootstrap-versus-compact loading and pulls directives or tactics on-demand to maintain doctrine-aware agent context.

To iterate on implement and review steps without advancing the step ID, the runtime control loop keeps the same step_id while returning different WP IDs. This continues until all WPs reach terminal or handoff lanes, ensuring WP-level iteration without step advancement.

A completed mission returning a step is a known workaround in the runtime control loop. You handle it by checking the returned decision kind; when the kind is terminal, you run final acceptance instead of executing the prompt as a standard step.

When guard conditions fail or the prompt file is null during workflow automation, the runtime control loop requires strict handling by interpreting the returned decision kind as blocked. You must resolve the underlying blockers before the mission can advance safely.

Mission orchestration via the runtime control loop requires no external dependencies. It operates independently using the canonical entry point to execute mission steps, handle decision kinds, and manage doctrine-scoped context loading during stalled or blocked workflows.