OpenCode JOC (Joint Operations Center)

Hub-driven multi-agent orchestration for OpenCode — interactive menus, not memorization.

Named after the military field command post. Yes, another milspec acronym in dev tooling. We know.

The problem

OpenCode configs pile up capabilities faster than anyone can keep track of. 64 skills. 29 agents. 6 commands. At some point you stop discovering things and start hoping they exist. JOC's answer: four hub commands that give you a menu instead of a scavenger hunt.

The four hubs

Hub	What it's for	Example
/ideation	Plan and research before you paint yourself into a corner	/ideation plan sprint 5
/orchestrate	Pick an execution pattern and let it run	/orchestrate ralph fix TS errors
/project	Git, reviews, PRs, tests, icons, changelogs	/project review
/harvest-context	Save things before the context window forgets them	/harvest-context session

No arguments? You get a menu. Know what you want? Skip it: /project commit, /orchestrate ultrawork.

How hubs work

1. No arguments —> interactive menu with a description for each option
2. With subcommand —> terse reminder, check for prior state, delegate to the skill or agent
3. After completion —> suggest the next hub (ideation -> orchestrate -> harvest-context, project -> harvest-context)

Three of the four hubs (ideation, orchestrate, harvest-context) save progress to .opencode/state/ so you can resume later. Project is stateless, which is probably for the best.

Hub subcommands

/ideation — plan before you build

Subcommand	Delegates to	What it does
plan	plan skill	Interview -> structured work plan (or direct/consensus mode)
refine	idea-refine skill	Diverge/converge on an idea
deep	deep-interview skill	Socratic interview with ambiguity gating
graph	graph-thinking skill	Map complex relationships as graphs
research	ccg skill	Multi-model consultation
ralplan	ralplan skill	Planner -> Architect -> Critic consensus loop

/orchestrate — pick a pattern, let it run

Subcommand	Delegates to	What it does
ralph	ralph skill	Persistent loop until verified complete
team	team skill	N coordinated agents on shared tasks
deep	deep-dive skill	Trace -> deep-interview pipeline
ccg	ccg skill	Multi-model synthesis
ultrawork	ultrawork skill	Maximum parallel execution
autopilot	autopilot skill	Autonomous from idea to code
sciomc	sciomc skill	Parallel scientist agents

/project — ship it

Subcommand	Delegates to	What it does
tests	create-tests command	Generate 8-type test suite
commit	conventional-commit skill	Conventional commit with emoji prefixes
stage	git-stage-thread command	Stage files from current thread
review	code-reviewer agent	Code review: smells, security, architecture
pr	pr command	Create, view, merge, manage PRs
gh	github-ops skill	Full GitHub CLI operations
optimize	optimize command	Performance, security, maintainability
icon	icon-generator skill	Web/PWA/UE icon assets from source image
organize	file-organizer skill	Find duplicates, suggest structures
analyze	analyze-patterns command	Code pattern and architecture analysis
changelog	changelog-generator skill	User-facing changelog from git history

/harvest-context — save things before the context window forgets them

Subcommand	Delegates to	What it does
session	inline	Extract decisions and patterns from the conversation
codebase	deepinit skill	Deep codebase documentation
skill	skill-creator skill	Create or update a workflow skill
agent	opencode-agent-creator skill	Create agent definitions
rule	inline	Codify a reusable rule or pattern
command	opencode-command-creator skill	Create custom slash commands
memory	remember + wiki	Classify and persist project knowledge
docs	Context7 MCP	Fetch up-to-date library docs
decompose	planner agent	Break requirements into implementable units
context	inline	Capture full session context

Magic keywords

Skip the menu with natural language triggers. They call the subcommand directly. Convenient, until you accidentally type "don't stop" in a normal sentence and ralph mode activates.

Keyword	Hub	Subcommand
"ralph", "don't stop"	orchestrate	ralph
"autopilot", "build me"	orchestrate	autopilot
"ultrawork", "ulw"	orchestrate	ultrawork
"deep interview"	ideation	deep
"plan"	ideation	plan
"cancel", "stop"	—	cancel active mode

Quick start

# Explore — menus guide you
/ideation                  # -> interactive menu
/orchestrate               # -> interactive menu
/project                   # -> interactive menu
/harvest-context           # -> interactive menu

# Go direct
/orchestrate ralph fix all TypeScript errors
/project commit
/project review
/ideation plan redesign the auth module
/harvest-context memory

What's inside

Agents (29)

Category	Agents
Implementation	@executor, @code-simplifier, @refactoring, @frontend-design
Architecture	@architect, @planner, @analyst, @deep-thinker, @designer
Review	@code-reviewer, @security-reviewer, @critic, @verifier, @qa-tester, @test-engineer
Research	@scientist, @explore, @writer, @document-specialist
Workflow	@tracer, @git-master, @debugger, @config-orchestrator
Specialized	@effort-estimator, @prompt-simplifier, @skill-creator, @requirements-analyzer, @commit-drafter

Skills (64)

Category	Key skills
Execution modes	ralph, autopilot, ultrawork, team, ultraqa, cancel
Planning	plan, ralplan, deep-interview, deep-dive, sciomc
Quality	ai-slop-cleaner, self-improve, verify, visual-verdict
Development	deepinit, changelog-generator, skillify, learner
Setup	init-project, joc-doctor, joc-reference, mcp-setup
Hubs	ideation, orchestrate, project, harvest-context

Yes, 64 skills. Nobody has them all memorized. That's the point of the hubs.

Commands (6)

Command	Purpose
/init-project	Initialize or refresh project config + docs
/ideation	Planning and research hub
/orchestrate	Execution hub
/project	Project operations hub
/harvest-context	Context and artifact hub
/skill	Manage workflow skills

Tools (10)

Tool	Description
loadSkill	Load skill content and metadata
runSkillScript	Execute skill scripts
agentContext	Manage project memory and notepad
listAgents	List and filter agents
modeState	Manage execution mode state
taskTodos	Create and manage task lists
artifacts	Save and load skill artifacts
getSessionID	Get current session ID
saveCommitMessage	Save commit message
getCommitMessage	Retrieve saved commit message

Installation

# Install globally
curl -fsSL https://raw.githubusercontent.com/Thomashighbaugh/opencode/main/install.sh | bash -s -- --global

# Or install per-project
curl -fsSL https://raw.githubusercontent.com/Thomashighbaugh/opencode/main/install.sh | bash

Model configuration

Model	Context	Output	Best for
glm-5.1:cloud	202K	131K	General purpose
kimi-k2.5:cloud	262K	262K	Extended context
minimax-m2.7:cloud	205K	128K	High performance
qwen3.5:cloud	262K	32K	Long documents

Directory structure

~/.config/opencode/
├── opencode.jsonc       # Main configuration
├── AGENTS.md            # Project-level instructions
├── agents/              # 29 agent definitions
├── skills/              # 64 workflow skills
├── commands/            # 6 hub commands + init-project
├── tools/               # 10 TypeScript tools
├── plugins/             # Hook system plugin
├── rules/               # 9 shared rules
├── state/               # Session state (gitignored)
├── templates/           # File templates
└── docs/                # Documentation

Documentation

File	Description
installation.md	Installation methods, configuration, upgrading
execution-modes.md	Ralph, autopilot, ultrawork, team, ultraqa
agents.md	All 29 agents with descriptions and usage
skills.md	All 64 skills organized by category
commands.md	All 6 commands with examples
tools.md	TypeScript tools API reference
model-configuration.md	Model setup and routing
state-management.md	Session state persistence
plugin-system.md	Hook system and keyword detection
path-conventions.md	File and directory structure

License

MIT