- Get started
- Best practices
- Working effectively
- Give your agent project context
- Reviewing and validating output
- Capabilities
- Skills
- CLI
- Examples
- Examples overview
- Recipe: build an agent
- Recipe: automate with RPA
- Recipe: enter data into a web app
- Recipe: extract data from a desktop app
- Recipe: refactor and test a workflow
- Recipe: a queue-based process with REFramework
- Recipe: build a Maestro Flow
- Recipe: coded app and API workflow
- Recipe: verify a release with Test Manager
- Recipe: extract data from documents with IXP
- Advanced
- Help
Every session, your coding agent starts fresh. Without a standing source of project conventions, the same instructions get repeated, and the agent drifts when one is forgotten. A context file solves this: a file the agent reads automatically at the start of a session, holding the rules and conventions it should follow every time.
Why a context file
Re-typing your conventions in every prompt is slow and easy to get wrong. A context file makes those instructions persistent and consistent, so the agent follows your house style by default instead of guessing.
AGENTS.md, the common convention
AGENTS.md is an emerging cross-tool convention: a Markdown file at the root of your project that coding agents read for project instructions. Some agents look for their own filename instead, but the idea is the same regardless of the exact name. The agent's own documentation lists the filename and location it expects.
Global and project scope
Most agents support two levels of context:
- Global holds your personal standing preferences, applied across every project, such as your default UiPath organization and tenant, or your preferred style.
- Project holds conventions specific to one project, kept in the project's context file so they travel with the repository and the whole team benefits.
Organization-wide rules belong in the global context; project-specific rules belong in the project file.
What to put in it
A useful context file is short and concrete. For UiPath work, useful contents include:
- The UiPath skills for this project, so the agent reaches for the right one (for example,
uipath-rpafor Robotic Process Automation (RPA) work, oruipath-maestro-flowfor process orchestration). - A standing quality rule, such as "always run the Workflow Analyzer and resolve all problems before declaring the work done."
- Preferred and pinned packages, and the activities to favor or avoid (for example, native activities over Invoke Code).
- Project conventions: naming, structure, and the patterns your team follows.
- Explicit constraints the agent must respect, such as not overwriting recorded UI workflows or existing configuration files.
A short AGENTS.md for a UiPath project might look like this:
# Project conventions
- Use the `uipath-rpa` skill for automation work in this project.
- Always run the Workflow Analyzer (Analyze Project) and resolve every problem before calling work done.
- Prefer native activities over Invoke Code; keep UI activities inside a Use Application/Browser scope.
- Pin package versions in project.json; do not upgrade packages without asking.
# Project conventions
- Use the `uipath-rpa` skill for automation work in this project.
- Always run the Workflow Analyzer (Analyze Project) and resolve every problem before calling work done.
- Prefer native activities over Invoke Code; keep UI activities inside a Use Application/Browser scope.
- Pin package versions in project.json; do not upgrade packages without asking.
For RPA-specific conventions worth capturing here, see RPA workflows.
Keeping it current
A context file is only as good as its contents. A file that drifts out of date quietly misleads the agent the same way a wrong comment misleads a developer, so it is worth pruning as your conventions change.