tedivm/opencode-config

Purpose and Runtime

This is a personal opencode CLI configuration repository, intended to populate ~/.config/opencode. It defines agent behavior, model/provider integration, reusable instruction skills, and custom prompt commands rather than an application service. The primary runtime is the OpenCode CLI, configured by opencode.jsonc, with persisted conversation state in dcp.jsonc.

Core Configuration Boundary

opencode.jsonc is the central integration point for the local vLLM OpenAI-compatible provider, the qwen3.6-27b model, LSP/formatting capabilities, Context7 MCP documentation lookup, and the @tarquinen/opencode-dcp plugin. AGENTS.md establishes global agent operating rules. Secrets and endpoint settings are deliberately external environment files rather than repository configuration.

Paths: opencode.jsonc, dcp.jsonc, AGENTS.md.

Agent Specialization

Role-specific agent prompts define delegated work patterns and boundaries: architecture design, repository research, and review. These are reusable subagent personas invoked by the OpenCode workflow, rather than executable processes.

Paths: agents/architect.md, agents/researcher.md, agents/reviewer.md.

Command Layer

Custom commands package recurring, user-invoked workflows: repository critique, OpenSpec proposal/spec drift analysis, GitHub Action upgrades, response compression, and architecture/design review. Commands orchestrate configured agents and skills through prompt instructions; they are the main interactive entry points beyond the base CLI.

Paths: commands/repo-critique.md, commands/openspec-review.md, commands/document-architecture.md.

Skill Library

Each skills/*/SKILL.md is a domain workflow contract that supplies procedural instructions, conventions, and sometimes supporting assets or reference material. Major domains include project bootstrapping, code exploration, command/skill authoring, Python scaffolding, OpenSpec setup, PR creation, GitHub operations, JSON processing, quality validation, and design-system work.

Paths: skills/bootstrapping-plan/SKILL.md, skills/exploring-code/SKILL.md, skills/robs-design/SKILL.md.

Skill Assets and Reference Knowledge

Some skills carry executable or static supporting material. Markdown conversion invokes a Python integration with Microsoft MarkItDown; the Rob’s theme skill supplies CSS tokens/layout/component assets; architecture/design skills include reference patterns for FastAPI, Celery, SQLAlchemy, Pydantic, and Typer. These resources constrain generated output but are not a shared application runtime.

Paths: skills/converting-to-markdown/scripts/markitdown-llm.py, skills/robs-theme/assets/variables.css, skills/robs-design/references/fastapi-routes.md.

Installation and Validation

make install installs snip, an output-compression dependency used by the configuration/workflows. Repository validation is implemented in Node.js and invoked by the Makefile and CI. GitHub Actions enforce formatting and validation on changes.

Paths: Makefile, scripts/validate.js, .github/workflows/validate.yml.

External Services

The configuration depends on a locally hosted vLLM server through an OpenAI-compatible API, Context7 via MCP for documentation retrieval, GitHub through gh-oriented skills/commands, and the DCP plugin for conversation persistence. Credentials and host data are expected from untracked files such as .llm_host, .openai_api_key, and .context7-api-key.