session-plan

You can break down session scope by generating a structured wave execution plan that converts agreed VCS issues into ordered waves. This plan maps roles to wave numbers, assigns file-scoped tasks to discovered agents, and establishes dependency ordering alongside measurable acceptance criteria.

Agent orchestration in multi-wave planning coordinates parallel execution by assigning specific files to designated agents. It uses complexity scoring and file-affinity grouping to prevent parallel-change conflicts, ensuring isolated execution across configured waves and agents-per-wave constraints.

Assign agents by discovering project and plugin agents, honoring explicit agent-mapping, and auto-matching tasks by domain. File-scoped task distribution uses grep and glob discovery with complexity sizing and deconfliction rules to prevent parallel-change conflicts during wave execution.

Yes, you can resume execution planning from a previous session state by reading STATE.md. The planning process surfaces historical learnings to size tasks and warn about risks, then updates VCS issues with planned wave status and comments for continuity.

Session planning works directly with VCS issues by integrating them into the wave decomposition and updating their status post-planning. It honors session-config constraints including agent isolation, enforcement rules, max-turns, and persistence to ensure safe parallel execution.

Use wave decomposition for feature, deep, and housekeeping engineering sessions where VCS issues and file scopes must be decomposed into ordered agents. It is necessary when parallel changes risk conflicts and ad-hoc coordination fails to manage complex dependency ordering.