Skills
Dark Factory includes nine slash commands that scaffold project artifacts. Skills are installed to
.claude/skills/ by godark init and invoked in Claude Code.
Workflow Sequence
These skills follow a natural progression — each builds on the output of the previous:
/godark-create-milestone /godark-create-planning-doc /godark-create-issues /godark-create-scenarios /godark-create-phase-overview /godark-create-milestone Create a phased roadmap and generate GitHub milestones Gathers context from your existing docs (CLAUDE.md, architecture, conventions), discusses goals with you, proposes a phased breakdown, then writes the roadmap and creates milestones via the GitHub CLI.
/godark-create-planning-doc Expand a roadmap phase into detailed planning specs Analyzes codebase impact (files changed, layers involved, caller count), explores the change surface, then writes detailed specs with acceptance criteria and test cases. Enforces task sizing rules — max 5 acceptance criteria and 7 test cases per task. Oversized tasks are split into child issues.
/godark-create-issues Convert planning doc into GitHub issues Reads the planning doc, creates issues via the GitHub CLI, applies phase labels, assigns to the correct milestone, and updates the planning doc and roadmap with issue numbers. Avoids creating duplicates.
/godark-create-scenarios Generate scenario spec files for testing Creates markdown scenario specs that the functional reviewer validates against. Each spec has setup instructions and test cases with expected outcomes. Follows the canonical format from the spec generator prompt. Skips existing specs to prevent overwrites.
/godark-create-phase-overview Generate overview document for a completed phase Reads the actual implementation to document what was built — not what was planned. Grounds examples in real code (config names, CLI flags, struct names) with realistic scenarios and narrative context. Target length: 150-300 lines.
Utility Skills
These can be run at any time, independent of the workflow sequence:
/godark-define-architecture Define or update architecture layer definitions Scans your codebase for packages and imports, detects circular dependencies, and proposes idiomatic layers for your language/framework. Layers define which packages may depend on which — validated by godark vet architecture. The JSON schema defines name, description, paths, may_depend_on, and must_not_depend_on for each layer.
/godark-define-conventions Define or update coding conventions and patterns Analyzes 10-20 source files for patterns in use, then collaborates with you to define conventions for error handling, logging, test style, naming, and dependency injection. Applies an agent-friendliness filter — flags anti-patterns like heavy code generation, convention-over-configuration magic, and implicit behavior that impede autonomous agents.
/godark-configure-project Auto-detect and populate godark.yaml configuration Detects modules (go.mod, package.json, pubspec.yaml), code generation tools (sqlc, gqlgen, mockery, protobuf), environment variables (.env.example, CI workflows), CI checks (GitHub Actions jobs), and Docker Compose usage. Proposes language-specific defaults and confirms before writing.
/godark-harness-claude-md Compress and reorganize CLAUDE.md as a directory of pointers Classifies existing CLAUDE.md content as duplicated, misplaced, routing, or unique. Moves misplaced content to the right subordinate doc (architecture → docs/architecture.md, conventions → docs/conventions.md, build commands → godark.yaml). Removes duplicated content. Never silently deletes — you decide each move.