Skip to content

sigmastudioz/domain-driven-design

v1.0.7MIT

Principal Driven Domain Design: principal-level domain architecture, reusable engines, module gap tracing, and explicit agent handoffs.

GitHubOpen Website

Principal Driven Domain Design

By Lex from SigmaStudioz. Version 1.0.7, MIT.

Principal-level domain-driven design for AI coding agents.

Build features that fit the codebase—and trace the connections that make them work. An engineering baseline for application, API, worker and monorepo changes. Group code by its owning domain, then by function; keep services and shared modules at the narrowest useful scope. Reuse an owned engine when real consumers share a cohesive algorithm. Trace modules through callers, dependencies, consumers and failures, and fix gaps required by the authorized task.

The instruction file has 498 lines; the complete skill folder has 597 lines, including metadata and the optional Python helper. Small fixes stay small; read-only reviews stay read-only. The baseline does not authorize extra agents, unrelated restructuring or deployment.

What it helps agents do

  • Put code with the domain that owns it, then group services and shared modules by function.
  • Reuse cohesive engines across real consumers without creating a catch-all framework.
  • Follow callers, dependencies, consumers and failure paths to find and repair required gaps.
  • Give authorized teammates explicit ownership, contracts and evidence for integration.

Setup and handoffs

Install this plugin from the SigmaStudioz catalog. For global automatic use without trigger words, follow the one-time native setup. Marketplace discovery alone does not install global instructions or enable lifecycle hooks.

Pass an accessible baseline, workspace/revision, ownership, contracts, current state, evidence, trace obligations, checks and expected return format to already-authorized teammates. Verify the integrated result. Task briefs and pseudocode are proposals, not proof of algorithm correctness. User teammate limits override optional workflow steps, including extra review agents.

The optional Claude helper checks handoff structure, places the full baseline before the proposed task and enforces explicit Use N teammates/agents/workers limits from native user history. It counts pending dispatches, ignores hook-rejected attempts and permits resuming an existing worker. It does not infer authorization, grant tools or prove semantic correctness. Missing history or other constraint wording leaves budget enforcement to the coordinator. Requires Python 3.

Validation

Version 1.0.7 adds the Claude display name and discovery keywords. Version 1.0.6 introduced the public title and listing metadata. Its engineering instructions are unchanged from 1.0.5 apart from the title; the helper is byte-identical. The native coding evidence below was collected with 1.0.5. No fresh 1.0.6 native coding run is claimed.

Fresh Claude Code 1.0.5 test: 16/16 external feature checks and 2/2 original regressions pass, with exactly one implementation teammate. The parent integrated the isolated worktree result, ran 20 local tests, reviewed authority paths itself and fixed a forged-tenant field bug. Compound identity uses a JSON tuple instead of a separator, including NUL.

The test used normal global configuration with the optional handoff helper and an ordinary prompt that did not name the skill. The actual worker received the complete baseline before the assignment. Guard tests: 21; installer tests: 14. See the validation receipt. Earlier failed Claude runs remain recorded. The earlier passing Cursor CLI evidence is tied to its original baseline; no new Cursor desktop or 1.0.5 coding result is implied.

Production performance, durable payment concurrency and remote/cloud adherence are outside this small fixture. Local installation does not establish universal behavior across models or hosts.

Example

An ordinary request such as “Add invoice refunds to the API and retry worker” should make the agent inspect the billing domain, reuse a shared billing service, wire both consumers, and trace retry, authorization and persistence paths. Automatic application requires the native setup above.

Data and support

No publisher-hosted service, telemetry or automatic usage upload is included. The optional helper reads the host-provided transcript locally; the agent provider handles prompts according to user settings. See data handling. Support: GitHub Issues. Do not post secrets, private code or raw session logs. The MIT copyright notice credits Lex from SigmaStudioz.