# Repository Instructions Clarify Prompt Files' Roles > Source: > Published: 2026-07-07 04:06:27+00:00 # Repository Instructions Clarify Prompt Files' Roles Microsoft's **.NET Blog** clarified in a **September 2025** post (surfaced again in recent developer discussion) how **GitHub Copilot**'s two customization layers differ: repository-wide copilot-instructions.md files apply automatically to every request, while *.prompt.md files are reusable, task-specific templates invoked on demand, such as a code-review checklist. Microsoft's VS Code documentation confirms the same split and adds a third layer, agent skills, for bundled multi-step workflows. For developers and ML practitioners wiring Copilot into CI/CD and code-review pipelines, treating persistent conventions and one-off task prompts as separate artifacts reduces duplicated boilerplate and makes prompt-driven automation easier to version and audit. Distinguishing **repository instructions** from **prompt files** is a small but practical governance pattern that helps teams maintain stable, auditable Copilot behavior while keeping task-level workflows modular. Treating project conventions and one-off task prompts as separate artifacts reduces repeated boilerplate, simplifies updates to coding standards, and makes prompt-driven automation easier to test and version. ### What happened According to the **Microsoft .NET Blog**, instruction files, often named copilot-instructions.md, "define the rules, coding standards, or guidelines Copilot should follow" and are placed in the .github folder. The same post notes teams can also add file- or scenario-specific instruction files (for example title.instructions.md) that apply only to certain requests. Microsoft's Visual Studio Code documentation describes the same distinction: instructions apply automatically to every request, while prompt files are saved, reusable prompts invoked explicitly as a slash command, such as /scaffold-component or /prep-pr. VS Code's documentation adds a third customization layer, agent skills, for bundling multi-step, repeatable workflows with scripts and examples. ### Technical context From an engineering perspective, this separation maps to two different concerns: persistent policy or configuration versus executable prompt templates. Repository instructions act as global context that should remain stable across conversations; prompt files encapsulate the variable, task-level prompt an engineer runs for a specific action. That separation mirrors common software patterns that isolate configuration from behavior. ### For practitioners For developers and ML practitioners embedding Copilot into CI/CD and code-review workflows, keeping repository-level instructions and task-specific prompt files as separate artifacts improves repeatability and reduces drift in automated suggestions. Naming and placement conventions, such as .github/copilot-instructions.md versus .github/prompts/*.prompt.md, determine how predictably these files load and how easy the resulting automation is to version and audit. ### What to watch Watch for how teams adopt these naming and placement conventions, and whether CI or code-review tooling begins to reference prompt files directly rather than pasting instructions inline. Adoption patterns will determine how easy it becomes to version, test, and audit prompt-driven automation across repositories. ## Key Points - 1Repository instructions load automatically and set persistent, repository-wide Copilot coding standards. - 2Prompt files are reusable, task-specific templates invoked on demand, such as a code-review checklist. - 3Separating the two artifact types reduces duplicated boilerplate and improves auditability of Copilot automation. ## Scoring Rationale A practical, product-level clarification for developers and ML practitioners using Copilot for automated code review and CI/CD workflows. Useful governance guidance but incremental, not a model or platform breakthrough; corroborated by two independent Microsoft sources (.NET Blog and official VS Code docs). ## Sources Public references used for this report. Practice interview problems based on real data 1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with. [Try 250 free problems](/problems)