from hand-compute37
A technique skill that guides agents to execute system flows step-by-step with explicit state to find race conditions and state bugs.
Hand-Compute trains an agent to become a human 'computer': it executes a system manually, writing explicit actor state at each transition to reveal races, stale snapshots, and sequencing bugs. It provides structured shapes for new work, debugging, and scoping features.
Use for debugging regressions, race conditions, client/server sequencing bugs, recovery paths, or when scoping a feature onto an existing stateful system. Trigger on intents like "why is this broken", "walk through the flow", or "stress test this plan".
Ideal for code-capable assistants and reviewers (Copilot/GitHub-code assistants, Codex-style agents) that can read source files and synthesize concrete state traces.
This skill has not been reviewed by our automated audit pipeline yet.
