Agent Development
Norm-Skill provides a task index for CLI usage and language rules. Library Skills belong to the repositories that publish those libraries. Use the repository Skill publication contract to locate an immutable release Skill after resolving a package through the Registry.
Start with the installed norm -h for CLI usage. Each command has norm <command> -h; refactoring types use norm refactor name -h. Help does not execute project operations.
Use language and APIs supported by the current toolchain. See Status for capability maturity and Agent Tooling Design for principles and plans. A capability in a source checkout does not necessarily exist in an official release.
See the executable self-extending Agent for continuous source evolution. It uses the Agent module, which builds on the protocol-only OpenAI module.
Learning and querying
Semantic Query provides project overviews, declaration search, strong reference selection, and context.
Semantic Refactoring Preview provides edit previews and static validation tied to document revisions.
- Language Tour: sequential learning with runnable examples;
- Language Reference: exact syntax and semantics;
- Design Principles: type, information, and layering boundaries;
- Standard Library: current module entry points;
- API Documentation Export: public declarations and docs from the semantic model;
- VS Code: completion, signatures, navigation, references, and renaming.
For type relationships, first consult value, class, and ref semantics, nullability and inference, functions, and error handling. Do not infer Norm behavior from other languages.
Verification
See Checks and Tests for static checks, focused tests, and machine output. See the Testing API for test declarations and associations. The Norm test index describes project-local runnable examples and their verification commands.
A verification claim must state what actually ran. Successful compilation proves static constraints only; passing tests prove only their executed assertions. Revalidate the corresponding sources after changes.
See Agent Task Benchmark for reusable task preparation, independent acceptance, and measurement boundaries.