Nimi Coding
Nimi Coding gives AI hosts and repository tools deterministic access to project-owned canonical authority. It formats, checks, queries, relates, and reviews authority units; it is not an AI agent, planner, code generator, approval workflow, or product-spec generator.
This page is an explanatory projection of the exact @nimiplatform/nimi-coding@0.6.1 package pinned by this workspace. The package CLI and its bundled README are the operational reference for command behavior.
Truth Boundaries
| Path | Role |
|---|---|
.nimi/spec/**/*.authority.{yaml,md} | The only canonical product authority |
.nimi/config/** | Repository-owned Nimi Coding host configuration |
.nimi/methodology/authority-authoring.yaml | Managed authoring instructions |
.nimi/local/** | Ignored local verification and derived output; never authority |
config/** | Product, generator, and host implementation inputs; not Nimi Coding host configuration |
Nimi Coding does not turn documentation, fixtures, generated output, audit results, or config/** projections into product authority.
Bounded Code Reads
code context derives a bounded, outbound TypeScript or TSX context from one explicit source file, top-level symbol, and tsconfig. code authority locates optional exact @nimi-authority and @nimi-deprecated markers in tracked current TypeScript, TSX, Go, Python, and Rust source, or reads those links back from one source file. These reserved standalone physical lines are located without proving their language comment context.
Both operations are request-local and read-only. The AI host decides when to call them. They do not add a task preflight, change host task state, evaluate unannotated code, or prove that implementation conforms to authority.
Current Workflow
When an exact authority ID is not already known, use bounded discovery and then let project authority or its responsible owner select the exact unit. Before changing authority, obtain its declared context. After editing, format every changed container, run the complete-root authority check, and use semantic diff and impact with explicit byte budgets.
For redesigns that may leave unchanged active survivors outside authored relations, authority change-candidates recalls the complete union of only the explicitly requested deterministic channels. It does not decide relevance, conflict, retirement, or conformance. authority audit evaluates only the project's configured graph and exact lexical bindings; a clear result does not evaluate business semantics or implementation alignment.
Nimi's repository instructions in AGENTS.md and .nimi/methodology/authority-authoring.yaml define the required authoring rules. Their managed text guides AI hosts; it does not intercept prompts or own the host task lifecycle.