blueprint

To break down a large software refactor, you convert a high-level objective into a step-by-step executable construction plan with dependency graphs and parallelism detection. This produces per-step self-contained context briefs that independent cold-start agents can execute simultaneously.

A multi-agent construction plan is an ordered, executable sequence of steps for multi-session engineering projects. It assigns dependency edges, detects parallelizable work, and sets rollback strategies so fresh agents can run steps independently without context loss.

Creating self-contained context briefs involves splitting objectives into one-PR-sized steps and emitting Markdown plan files with verification commands and exit criteria. This ensures any fresh agent can run a specific step cold without prior project context.

Branch, PR, and CI guidance for multi-PR migrations is supported when git and gh are available. The process performs pre-flight checks for git, gh auth, remote, and default branch to structure your rollout correctly.

The best way to validate a multi-PR plan against anti-patterns is to run an adversarial review gate. A strongest-model review pass checks the plan against an anti-pattern catalog and fixes critical findings before finalizing.

Managing plan mutation and registration for parallel agent implementations uses a plan mutation and registration protocol. It saves the emitted Markdown plan files and updates memory indexes to coordinate parallel workstreams and prevent coordination errors.