DEV_COMPANION

Dev Companion

The dev companion provides AI-powered delivery workflow assistance. For full details, see docs/DEV_COMPANION.md.

Layers

Layer	Skill	Purpose
L1	dev-assistant	Orchestrator + repo inspection (every repo)
L2	dev-companion	General dev companion framing for client work

Workflow skills (workflow-generic-project) drive phases and gates; tool skills own CLI procedures.

How it works

The dev companion follows a WHAT/HOW pattern:

• WHAT skills define delivery phases: plan → implement → review → PR
• HOW skills handle tool-specific CLIs (GitHub, GitLab, dbt, etc.)

graph LR
    A[Dev Companion WHAT] --> B[Plan Phase]
    B --> C[Implement Phase]
    C --> D[Review Phase]
    D --> E[PR Phase]
    E --> F[Tool Skills HOW]

Loading

Optional background runner

For autonomous processing, the dev companion includes a queue/worker system:

dots-devcompanion enqueue <id>    # queue a job
dots-devcompanion run-once        # process next job
dots-devcompanion status          # check queue
dots-devcompanion done <id>       # mark complete

Note

IDE-first workflows are the default. The background runner is optional.

LLM integration

The runner includes a provider-agnostic LLM layer supporting multiple backends:

Provider	Type	Config required
OpenCode (big-pickle)	Free, local	None (zero-config)
Ollama	Local GPU	OLLAMA_HOST
Claude	Cloud	ANTHROPIC_API_KEY
OpenAI	Cloud	OPENAI_API_KEY

See LLM Providers for full configuration.

See also

• AI Overview — the broader AI layer
• LLM Providers — provider priority and setup
• Skills System — how skills are managed