diff-first-review

Use when minimising GitHub Copilot AI credits, selecting a mode or model, reducing tool overhead, or deciding whether to start a fresh session.

Use when repository structure, architecture, documentation, conventions, or project context must be discovered before a change.

Use when changing task, topic, model, reasoning effort, agent, or tool set, or when accumulated conversation context is no longer relevant.

Use when only specific fields or records are required from JSON, YAML, XML, CSV, API responses, manifests, lock files, or other structured data.

Use when locating symbol definitions, references, implementations, call sites, or structurally similar code without reading broad file sets.

Use when creating, reviewing, splitting, or optimising Copilot instructions, agents, prompts, or skills for context cost.

Use when builds, tests, linters, logs, Git commands, or other terminal operations may return verbose output.

Use when a focused test, build, lint, type-check, or tool command fails during a task and the next repair is unclear.

Use when creating a project/repository, selecting its licence or default branch, merging branches, or pulling changes.

Use when the relevant files are known.

Enforces 100% template fidelity for GitHub markdown files (README.md, ARCHITECTURE.md, CONTRIBUTING.md, CODE_OF_CONDUCT.md, SECURITY.md, PRIVACY.md…

Use before changing code when test, build, lint, or type-check baseline status is unknown and pre-existing failures could cause rework.