@tmidev/hyper-opencode

OpenCode plugin for Agent Hyper.

Install

npm install @tmidev/hyper-opencode

Fast Setup

npx @tmidev/hyper-opencode init

Current package version: 0.0.5

init does three things:

1. Chooses config scope (global or project, default: global)
2. Adds plugin entry to OpenCode config
3. Writes ~/.config/opencode/agent_hyper.json with full built-in agent config (skills, autoLoadSkills, temperatures, and flags)
4. Copies copilot-auth.ts template to the selected scope:
   • project scope: <project>/.opencode/plugin/copilot-auth.ts
   • global scope: ~/.config/opencode/plugin/copilot-auth.ts

When agent_hyper.json already exists, init preserves existing per-agent model values and only refreshes other config fields. Use --model-* flags to explicitly override preserved models.

Config search paths:

• Project: <project>/opencode.jsonc then <project>/opencode.json
• Global: ~/.config/opencode/opencode.jsonc then ~/.config/opencode/opencode.json

Useful Flags

# no file writes
npx @tmidev/hyper-opencode init --dry-run --non-interactive

# force project scope
npx @tmidev/hyper-opencode init --non-interactive --yes --opencode-config-scope project

# set custom models
npx @tmidev/hyper-opencode init --non-interactive --yes \
  --model-hyper-maestro github-copilot/gpt-5.4 \
  --model-architect-lead github-copilot/gpt-5.3-codex

Planning-mode delegation

• Delegate to Scout when research is needed and you cannot name the file path upfront.
• Read-only exploration is allowed during planning; don't implement until execution.
• In planning mode, "don't execute" means don't implement.
• Canonical delegation threshold: delegate when work spans 2+ files, unclear ownership, or external systems.

Custom Derived Subagents

{
  "agents": {
    "forager-executor": {
      "variant": "medium"
    },
    "quality-guardian": {
      "model": "github-copilot/gpt-5.2-codex"
    }
  },
  "customAgents": {
    "forager-ui": {
      "baseAgent": "forager-executor",
      "description": "Use for UI-heavy implementation tasks.",
      "model": "anthropic/claude-sonnet-4-20250514",
      "temperature": 0.2,
      "variant": "high"
    },
    "reviewer-security": {
      "baseAgent": "quality-guardian",
      "description": "Use for security-focused review passes."
    },
    "general-helper": {
      "baseAgent": "general",
      "description": "Use for generic utility and glue tasks when OpenCode general is available."
    }
  }
}

Key Tools

• hyper_feature_create, hyper_feature_complete
• hyper_plan_write, hyper_plan_read, hyper_plan_approve
• hyper_tasks_sync, hyper_task_create, hyper_task_update
• hyper_task_start, hyper_task_resume, hyper_task_done
• hyper_context_write, hyper_status, hyper_skill

hyper_feature_complete now auto-runs an AGENTS.md sync proposal (same as hyper_agents_md({ action: "sync" })) and returns the proposed diff in the tool output. Changes are never auto-applied; use hyper_agents_md({ action: "apply", content: "..." }) after review.

Agents

• hyper-maestro
• architect-lead
• swarm-lead
• scout-analyst
• forager-executor
• quality-guardian

Built-in Skills

• brainstorming
• writing-plans
• executing-plans
• dispatching-parallel-agents
• parallel-exploration
• test-driven-development
• systematic-debugging
• using-skills
• verification-before-completion
• code-reviewer
• docker-mastery
• agents-md-mastery

Optional MCP IDs

• websearch
• context7
• grep_app
• ast_grep

Minimal Config Example

{
  "$schema": "https://unpkg.com/@tmidev/hyper-opencode/schema/agent_hyper.schema.json",
  "agentMode": "unified",
  "enableToolsFor": [],
  "disableSkills": [],
  "disableMcps": [],
  "sandbox": "none",
  "customAgents": {},
  "use_tdd": false,
  "use_paralell": true,
  "agents": {
    "hyper-maestro": {
      "model": "github-copilot/gpt-5.4",
      "temperature": 0.5,
      "skills": ["brainstorming", "writing-plans", "dispatching-parallel-agents", "executing-plans"],
      "autoLoadSkills": ["using-skills", "parallel-exploration"]
    }
  }
}

License

MIT with Commons Clause (see ../../LICENSE).