Your coding standards, applied to every line the AI writes
Instructions are short, always-on rules for a type of file — React components, C# services, Apex classes. The assistant applies them automatically, even to quick "vibe coding" requests, so code follows your standards without anyone having to ask.
Standards only work if someone remembers them
Quick requests like "add a page that lists orders" get the most common pattern online, not your team's — especially in fast, exploratory vibe coding.
Reviewers repeat "use the repository pattern", "no secrets in config", "parameterise that query" on pull request after pull request.
Teams cram every rule for every stack into one file. The assistant reads all of it on every request and the important rules get lost.
Short rules, scoped to the files they apply to
Each instructions file declares which files it covers (for example **/*.tsx). The assistant loads it only when it works on matching files — nothing to invoke, nothing to remember.
- 1You edit or create a file
For example a React component, a .NET controller or an Apex trigger.
- 2Matching rules load automatically
Copilot uses
applyTo; Claude Code uses the same globs as rulepaths. Other stacks' rules stay out of the way. - 3Code follows your standards
The assistant writes to the rules — and points to the full skill when a task needs more depth.
What teams get
Rules apply automatically to matching files. Developers get the standard without knowing it exists.
Quick, casual prompts still produce code that follows team conventions, security rules and patterns.
The things reviewers say every week are written down once and applied before the pull request is opened.
The secure-coding rules cover every file: no secrets in code, parameterised queries, authorisation checks, no PII in logs.
React rules for components, .NET rules for C#, Apex rules for triggers — scoped, so the context stays small and relevant.
One source file; the Claude Code rule is generated from it, so teams on either tool follow the same standards.
Every repository that installs the instructions follows the same baseline, which makes code easier to move between.
When a framework or standard changes, one pull request updates the rules for every team.
Same request, with and without instructions
Prompt: Add an endpoint that returns a customer by ID.
Without
- Builds SQL with string concatenation
- No authorisation check on who may read the customer
- Logs the full customer record, including personal data
- Returns stack traces to the caller on errors
- A reviewer has to catch all of it
With .NET + Secure coding instructions
- Uses parameterised queries through the data layer
- Checks the caller may access that customer
- Logs identifiers only, never personal data
- Returns a standard problem response without internals
- Follows the project structure the team agreed
Illustrative example based on the .NET and secure coding instructions. Results depend on the tool, the model and the codebase.
Instructions, skills or agents?
| Use | When | Example |
|---|---|---|
| Instructions | Rules that should apply every time a file type is touched | Naming, patterns, security basics for C# files |
| Skills | A deeper playbook for a specific task, loaded when relevant | Designing a .NET architecture, writing a test strategy |
| Agents | A repeatable workflow with its own tools and output format | Planning a work item, reviewing a pull request |
How to show it's working
Measure a baseline before rollout, then compare after a few sprints. These are the signals that move first — track them per team, not per person.
| Measure | Why it matters |
|---|---|
| Repeated review comments per PR | The clearest sign the rules are being applied. |
| Security findings in AI-assisted changes | The secure-coding baseline should reduce routine findings. |
| Lint and type errors in first commits | Standards applied at write-time mean fewer fix-up commits. |
| Repositories with instructions installed | Coverage across teams — the baseline only helps where it is installed. |
Useful at every level of the team
Write to the standard without memorising it — including during quick prototyping.
Stop repeating the same comments and focus on logic and design.
A baseline that reaches every repository and every assistant.
The right rules load automatically for whichever code they touch.
Pick the instructions your team needs
Each item in the catalog has copy buttons and install steps for GitHub Copilot and Claude Code.