---

Why

OpenCode already has powerful automation plugins. Many of them are built for advanced orchestration: multiple agents, metrics, scopes, model maps, verification plans.

oh-my-goal is intentionally smaller.

/goal fix the flaky login test

That is the interface. The plugin stores the goal, injects a continuation prompt whenever the session goes idle, and keeps the agent moving until there is a concrete stop condition.

Features

• One-line autonomy: start with /goal <objective>.
• Project-local state: stores goal state in .opencode/goal.json.
• Continuation loop: uses session.idle to push the agent into the next concrete action.
• Completion detection: watches for GOAL_ACHIEVED: and GOAL_BLOCKED: markers.
• Budget guardrails: tracks token usage and warns near the configured budget.
• Compaction-safe context: preserves goal state through OpenCode session compaction.
• No model orchestration: plays nicely beside other OpenCode tools because it only manages goal continuation.

Install

npm i -g oh-my-goal

Add it to opencode.json:

{
  "plugin": ["oh-my-goal"]
}

For local use before publishing, build the repo and run OpenCode from this project. The repository includes:

• .opencode/plugins/oh-my-goal.js to load the built plugin locally
• .opencode/commands/goal.md so the TUI registers /goal

Usage

Start a goal:

/goal fix the login redirect bug

Check or control the active goal:

/goal status
/goal pause
/goal resume
/goal clear

Override the default token budget:

/goal raise payment module test coverage to 80% --token-budget 100000

Commands

Command	Behavior
/goal <objective>	Create a new goal and start the loop
/goal status	Show current objective, status, budget, and iteration
/goal pause	Pause the loop without deleting goal state
/goal resume	Resume a paused goal
/goal clear	Delete .opencode/goal.json and stop the loop

How It Works

The plugin is deliberately simple:

/goal <objective>
  -> write .opencode/goal.json
  -> on session.idle, inject continuation prompt
  -> on message.updated, scan for stop markers and token usage
  -> preserve goal context during compaction

When a goal is pursuing, each idle event injects an active-goal check into the session. The agent must concretely decide whether the original objective is complete. If not, it continues immediately.

The loop stops when:

• The assistant emits GOAL_ACHIEVED: <summary>
• The assistant emits GOAL_BLOCKED: <reason>
• /goal pause or /goal clear is used
• The token budget or iteration limit is reached

State File

.opencode/goal.json follows this shape:

{
  "objective": "fix the login redirect bug",
  "status": "pursuing",
  "created_at": "2026-05-05T10:00:00.000Z",
  "updated_at": "2026-05-05T10:02:00.000Z",
  "token_budget": 50000,
  "tokens_used": 12400,
  "iteration": 3,
  "max_iterations": 100,
  "budget_warning_sent": false,
  "history": [
    {
      "iteration": 1,
      "summary": "Injected continuation prompt.",
      "status": "in_progress",
      "created_at": "2026-05-05T10:01:00.000Z"
    }
  ]
}

Development

npm ci
npm run check
npm run build

Quality gates:

• npm run typecheck
• npm run lint
• npm run format:check
• npm audit --omit=dev

Contributing

Issues and pull requests are welcome. See CONTRIBUTING.md, CODE_OF_CONDUCT.md, and SECURITY.md.

License

Apache-2.0. See LICENSE.