What problem does it solve? Coordinating the execution of many interdependent research plans within a phase is error-prone when done manually, especially when plans involve derivations, numerical simulations, data analysis, and LaTeX writing that must respect dependencies, conventions, and verification gates. ## Core Features & Use Cases - Wave-Based Parallel Execution: Discovers plans, analyzes dependencies, groups them into waves, and spawns executor subagents in parallel while detecting intra-wave dependency cycles and file-modification conflicts. - Phase Classification & Adaptation: Classifies phases as derivation, numerical, literature, paper-writing, formalism, analysis, or validation, then adapts context budgets, inter-wave checks, timeouts, and pre-execution agents accordingly. - Safety Gates & Checkpointing: Enforces convention locks, first-result gates, pre-fanout reviews, bounded execution segments, and git-tag wave checkpoints regardless of autonomy mode. - Use Case: A physics researcher with a planned phase containing five PLAN.md files runs the skill to execute them across two waves, with automatic verification, convention enforcement, and SUMMARY.md generation per plan. ## Quick Start Ask the AI to execute phase 1 with the gpd-execute-phase skill, optionally passing --gaps-only to run only gap-closure plans.
