Working with a coding agent
Precept ships two things for an agent writing tests in your project. The MCP server answers questions the project cannot answer for itself — which environment a run will use, why a green build found no tests, what a filter selects. The skills say how to write the tests: what a test class looks like, how a feature file and its bindings are shaped, where a setting goes and which registration reads it, how to run and filter, and how each module is used. A Copilot agent and an instructions file come with them.
They are published as autom3tion/precept-skills,
in the Agent Skills format that GitHub Copilot, Claude Code, Cursor and
the other agents reading SKILL.md files understand:
npx skills add autom3tion/precept-skills| Skill | Covers |
|---|---|
precept-tests |
[TestSuite]/[Test] classes, hooks, retries, outcomes and awaited assertions — see Writing tests and Assertions. |
precept-gherkin |
A .feature file, its Reqnroll bindings and the Reqnroll section — see Gherkin. |
precept-settings |
precept.json, overlays, PRECEPT_* variables, startup and your own settings — see Configuration. |
precept-running |
Running, listing and filtering, environments and artifacts — see Running and Filtering. |
precept-modules |
Rest, Browser, Db, TestData, Rpc and screenplay, one example each. |
precept-page-objects |
Page objects and components on IPageContext — see Dependency injection — and when to move to screenplay. |
precept-migrating |
Moving a suite from NUnit, xUnit, MSTest, SpecFlow, FluentAssertions, Selenium or RestSharp, with every old test accounted for. |
precept-pipelines |
Running the suite on Azure Pipelines with the shipped pipeline and job template, settings as PRECEPT_* variables, TRX and attachments, exit codes. |
GitHub Copilot#
Copy .github/agents and .github/instructions from the published repository into your own
.github/ folder, and put the skills under .github/skills/ — which is where the installer above
puts them for Copilot. The Precept agent declares the MCP server itself, launched with
dnx Precept.Mcp, so a project that adopts the agent needs no separate MCP configuration; it loads
the matching skill before it acts, reads precept.json and your IPreceptStartup before writing,
and finishes by listing and running the tests it changed.
Keeping them in step#
The skills name Precept's API — the entry points, the assertions, the settings keys, the order the environment is resolved in — so they are versioned with it: the published README says which Precept it describes, and every release republishes them. Take the version your project references, and if the MCP server reports a different one, prefer the project's — it is the code the agent will be compiling against.