@hung319/opencode-hive

@hung319/opencode-hive

From Vibe Coding to Hive Coding — The OpenCode plugin for plan-first, structured AI-assisted development with multi-agent orchestration.

Quick Start

Add @hung319/opencode-hive to your opencode.json:

{
  "plugin": ["@hung319/opencode-hive"]
}

That's it. On first load, Hive auto-installs:

• Agent Tools: @sparkleideas/agent-booster (52x faster code editing), @sparkleideas/memory (vector memory)
• CLI Tools: dora (SCIP code navigation), auto-cr-cmd (automated code review), btca (Bluetooth Classic Audio)
• Snip binary: 60-90% token reduction by filtering shell output

All tools fall back gracefully if installation fails — nothing breaks.

Configuration

Auto-generated at ~/.config/opencode/agent_hive.json:

{
  "agentMode": "unified",
  "agents": {
    "hive": { "model": "anthropic/claude-sonnet-4-20250514", "temperature": 0.5 }
  }
}

Option	Default	Description
agentMode	unified	dedicated splits planner + orchestrator into separate agents
disableSkills	[]	Globally hide skills from hive_skill()
disableMcps	[]	Globally disable MCP servers

Agent Models & Variants

{
  "agents": {
    "hive": { "model": "anthropic/claude-sonnet-4-20250514", "variant": "high" },
    "scout-researcher": { "model": "anthropic/claude-sonnet-4-20250514", "temperature": 0.5 },
    "forager-worker": { "model": "anthropic/claude-sonnet-4-20250514", "variant": "medium" },
    "hygienic-reviewer": { "model": "anthropic/claude-sonnet-4-20250514", "temperature": 0.3 }
  }
}

Variants map to model-specific settings in your opencode.json (e.g., Anthropic thinking budgets).

Custom Derived Subagents

Define derived agents from forager-worker or hygienic-reviewer:

{
  "agents": {
    "forager-worker": {
      "variant": "medium"
    },
    "hygienic-reviewer": {
      "model": "github-copilot/gpt-5.2-codex"
    }
  },
  "customAgents": {
    "forager-ui": {
      "baseAgent": "forager-worker",
      "description": "Use for UI-heavy implementation tasks.",
      "model": "anthropic/claude-sonnet-4-20250514",
      "temperature": 0.2,
      "variant": "high"
    },
    "reviewer-security": {
      "baseAgent": "hygienic-reviewer",
      "description": "Use for security-focused review passes."
    }
  }
}

Omitted fields (model, variant, temperature) inherit from the base agent. autoLoadSkills merges with base defaults, de-duplicates, and applies global disableSkills. Custom agent IDs cannot reuse built-in Hive names or reserved plugin IDs.

Skills

ID	Use Case
brainstorming	Explore requirements before implementation
writing-plans	Create detailed implementation plans
executing-plans	Execute tasks in batches with review checkpoints
dispatching-parallel-agents	Dispatch 2+ agents for independent work
test-driven-development	Red-green-refactor cycle
systematic-debugging	Root cause investigation before fixes
code-reviewer	Review changes against plan
verification-before-completion	Verify before claiming done
docker-mastery	Docker container debugging and optimization

Per-agent filtering: { "agents": { "forager-worker": { "skills": ["tdd", "verification-before-completion"] } } }

Auto-load skills: Use autoLoadSkills to inject skills into agent prompts at session start:

{
  "agents": {
    "hive": { "autoLoadSkills": ["parallel-exploration"] }
  }
}

Resolution order: Hive builtin → Project OpenCode → Global OpenCode → Project Claude → Global Claude.

MCP Servers

MCP	Tool	Requires
websearch	websearch_web_search_exa	EXA_API_KEY env
context7	context7_query-docs	None
grep_app	grep_app_searchGitHub	None

Tools

Category	Tools
Feature	hive_feature_create, hive_feature_complete
Plan	hive_plan_write, hive_plan_read, hive_plan_approve
Task	hive_tasks_sync, hive_task_create, hive_task_update
Worktree	hive_worktree_start, hive_worktree_create, hive_worktree_commit, hive_worktree_discard
Merge	hive_merge
Context	hive_context_write
Memory	hive_memory_*, hive_vector_*
Code	hive_code_edit, hive_lazy_edit, hive_booster_status
Other	hive_status, hive_skill, hive_agents_md

Planning-mode delegation

During planning, "don't execute" means "don't implement" (no code edits, no worktrees). Read-only exploration is explicitly allowed and encouraged, both via local tools and by delegating to Scout.

Canonical Delegation Threshold

• Delegate to Scout when you cannot name the file path upfront, expect to inspect 2+ files, or the question is open-ended ("how/where does X work?").
• Local read/grep/glob is acceptable only for a single known file and a bounded question.

Agent Booster (Ultra-Fast Code Editing)

Rust+WASM powered — 52x faster than Morph LLM, no API key needed.

hive_code_edit({
  path: "src/index.ts",
  oldContent: "const old = 'value';",
  newContent: "const new = 'updated';"
})

Vector Memory (Semantic Search)

HNSW indexing for semantic memory search:

hive_vector_add({
  content: "Use async/await instead of .then() chains",
  type: "learning",
  tags: ["javascript", "best-practice"]
})

hive_vector_search({ query: "async patterns", type: "learning", limit: 10 })

Prompt Budgeting

Hive bounds worker prompts to prevent context overflow. Defaults: 10 tasks, 2K chars per summary, 20K per context file, 60K total. Exceeded content is truncated with ...[truncated] markers and file path hints.

Troubleshooting

If a task gets stuck in a blocked-resume loop:

• Call hive_status() first
• If pending/in_progress: hive_worktree_start({ feature, task })
• Only use hive_worktree_create({ task, continueFrom: "blocked", decision }) when status is exactly blocked

License

MIT with Commons Clause — Free for personal and non-commercial use. See LICENSE for details.