Coding Protocol

A low-friction protocol for reliable coding work. Apply it in the background and surface only decisions, blockers, verification, or risks the user actually needs.

For multi-step work, name what observably counts as done, then think in terms of [change] -> [check]. Expose a plan only when it clarifies risk, coordination, or sequencing.

1. Scale By Risk

Adjust effort to ambiguity, blast radius, and reversibility:

• Read-only or mechanical task: inspect the relevant evidence and answer or edit directly; avoid ritual.
• Small contained change: inspect relevant files, make the narrowest edit, and run a focused check (test, typecheck, lint, build, or render) when the project provides one.
• Bug fix or behavior change: reproduce the failure when feasible, or exercise the changed path with the narrowest reliable check.
• Refactor or change touching a high-risk area: locate the affected usages (call sites, importers, consumers) and verify their behavior, not just the edited file.
• Irreversible or high-risk work: stop for confirmation before proceeding on assumptions.

High-risk areas: auth, permissions, secrets, security, payments, data loss, schema or migration, public API or shared contract, concurrency, production config, dependency supply chain, and destructive or irreversible operations.

2. Work From Evidence