dx

Owns the declaration. Use when the user asks to write, refine, tighten, prune, or restructure a declaration; to add or remove invariants; to promote assumptions to invariants; to introduce contracts; or to reclassify constraints between `## Invariants` and `## Unconstrained`. The architect is the only role permitted to modify `## Intent` and `## Invariants`.

Authoritative reference for the dx language. Use whenever you need to write, modify, or validate the structure of a declaration: which blocks are required, what each schema element means, and how to log assumptions correctly. Load this any time you are about to emit Markdown into a `.md` declaration or answer a spec-conformance question.

Generates or modifies imperative code to satisfy a dx declaration. Use when the user asks to "implement this spec", "write the code from `system.md`", "make the implementation match", or to fix code so it conforms. The implementer is the only role that touches imperative source under the spec's authority. It may add to `## Assumptions` and *only* `## Assumptions`; all other blocks of the declaration are read-only.

Verifies that an implementation satisfies the `## Contracts` block of a dx declaration via black-box testing, then classifies any failure as either an implementation bug or a spec gap. Use when the user asks to "verify the implementation", "run the contracts", "confirm conformance to `system.md`", or after the `implementer` declares its work complete. The judge writes no implementation code and modifies no blocks of the declaration except to flag findings; remediation is routed to the appropriate role.

Distill an existing imperative codebase into a dx declaration. Use when the user says "reverse-engineer this into a declaration", "extract a spec from these sources", "there's no declaration here yet — make one", or when you encounter a dx-style repo that lacks a declaration for an existing component. Produces a base declaration that the `architect` will then refine.

Use when working in a dx-managed repository (any directory containing one or more dx declaration files, or whose `AGENTS.md` references the dx paradigm). Routes the agent to the correct role-skill (archaeologist, architect, implementer, judge), enforces the prompt-first workflow, and prevents silent semantic drift between human intent and generated code. Read this before answering any task that involves reading, writing, or executing code governed by a declaration.

How to invoke the deterministic `dx` CLI (`lint`, `fmt`, `diff`, `export`) from inside any agent's event loop. Covers exit-code semantics, required flags, when each command is mandatory, and how to integrate the toolchain into the AGENTS.md verification loop. Load this whenever you are about to run `dx` as a subprocess or wire it into CI.