AI

AI Overview

The AI layer is the core of dots-ai — skills, agents, and the dev companion. For full details, see docs/AI_LAYER.md.

Directory structure (after install)

Path	Purpose
~/.local/share/dots-ai/skills/	Bundled skills (managed by chezmoi)
~/.local/share/dots-ai/skills-external/	External skills (JIRA, Confluence packs)
~/.local/share/dots-ai/mcp/	MCP provider templates
~/.local/share/dots-ai/dev-companion/	Dev companion runner + queue
~/.local/share/dots-ai/prompts/	Reusable internal prompts
~/.local/bin/dots-*	CLI helpers

Skills system

Skills are markdown documents (SKILL.md) that teach AI tools how to perform workflows. Each skill includes a skill.json manifest declaring compatibility with AI tools:

Tool	Skills directory
Claude Code	~/.claude/skills/
GitHub Copilot CLI	~/.copilot/skills/
Cursor	~/.cursor/skills/
OpenCode	~/.config/opencode/skills/
pi agent	~/.pi/agent/skills/

dots-skills sync reads each manifest and creates symlinks only for supported tools.

Note

See the Skills System page for full documentation on manifests, registry, and publishing.

AI Agents

13 specialized subagents are deployed to Claude Code, OpenCode, Cursor, and Windsurf:

Agent	Purpose
dev-assistant	Repo inspection and workflow orchestration
architect	System design and architecture
code-reviewer	Code quality and maintainability
security-reviewer	OWASP Top 10 and vulnerability detection
planner	Feature planning and task breakdown
tdd-guide	Test-driven development
refactor-cleaner	Dead code removal and simplification
build-error-resolver	Build, TypeScript, and CI failures
database-reviewer	PostgreSQL and query optimization
docs-lookup	Framework docs and API references
e2e-runner	Playwright end-to-end testing
performance-optimizer	Profiling and optimization
typescript-reviewer	TypeScript type safety

The Ralph Loop

dots-ai implements the Ralph Loop — a conceptual model for agentic AI:

Ralph concept	dots-ai implementation
Backing specifications	AGENTS.md templates deployed to each repo
Context engineering	Modular skills that prime each loop with domain context
Persistent memory	knowledge/ in the workspace instance
Fix the loop	workspace-knowledge-sync skill
Monolithic orchestrator	dev-assistant as single entry point
Forward mode	Dev companion driving delivery phases

See also

• Skills System — full skills documentation
• Dev Companion — companion layers and workflows
• LLM Providers — provider configuration