What problem does it solve? Physics calculations fail in subtle ways—sign errors, missing factors of 2, convention mismatches, numerical instabilities—that are hard to isolate and expensive to investigate in a single context window. This Skill orchestrates systematic root-cause diagnosis using dedicated investigation agents while keeping the main conversation lean. ## Core Features & Use Cases - Quick Triage Pattern Matching: Matches discrepancy signatures (factor of 2, sign flip, wrong power law, divergence) against a table of common physics error causes before full investigation. - Parallel Investigation Agents: Spawns one gpd-debugger subagent per issue with symptoms pre-filled, each with fresh context for reading derivations and testing hypotheses. - Persistent Debug Sessions: Writes investigation state to .gpd/debug/ session files so debugging survives context resets and supports checkpoint-based continuation. - Batch and Interactive Modes: Processes all gaps from VERIFICATION.md in parallel when called by verify-work, or runs a single interactive investigation loop with the user. - Use Case: A researcher's simulation shows energy drifting 1% over 1000 timesteps. The Skill gathers symptoms, spawns a debugger agent that identifies a forward Euler integrator replacing the symplectic Verlet scheme, and records the root cause in VERIFICATION.md for targeted fix planning. ## Quick Start Ask the AI to debug your physics issue by describing the expected result, the actual result, and how the discrepancy behaves, for example: "Debug why my energy conservation check drifts by 1% over 1000 timesteps."
