[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-81527":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":9,"language":10,"languages":9,"totalLinesOfCode":9,"stars":11,"forks":12,"watchers":13,"openIssues":14,"contributorsCount":14,"subscribersCount":14,"size":14,"stars1d":14,"stars7d":15,"stars30d":15,"stars90d":14,"forks30d":14,"starsTrendScore":14,"compositeScore":16,"rankGlobal":9,"rankLanguage":9,"license":17,"archived":18,"fork":18,"defaultBranch":19,"hasWiki":20,"hasPages":18,"topics":21,"createdAt":9,"pushedAt":9,"updatedAt":22,"readmeContent":23,"aiSummary":24,"trendingCount":14,"starSnapshotCount":14,"syncStatus":15,"lastSyncTime":25,"discoverSource":26},81527,"small-opencode-orchestrator","tempont\u002Fsmall-opencode-orchestrator","tempont","A lightweight implementation of an orchestrated multi-agent workflow, in which a central coordinator delegates tasks to specialized subagents to optimize token efficiency without compromising code generation quality.",null,"TypeScript",29,5,1,0,2,2.33,"MIT License",false,"main",true,[],"2026-06-12 02:04:16","# OpenCode Agent Orchestrator\n\nA lightweight OpenCode configuration that implements a practical orchestrator pattern for AI-assisted software development.\n\nThe goal is to get the benefits of a multi-agent workflow - planning, execution, review, and security checks - without turning the setup into a complex agent framework or adding excessive token overhead.\n\n## Why this exists\n\nSingle-agent workflows are often enough for small tasks, but they tend to become less reliable as work gets larger or more ambiguous. Common issues include:\n\n- Planning and implementation getting mixed together\n- Repeated context consuming unnecessary tokens\n- Reviews being skipped or performed inconsistently\n- Security, documentation, and testing concerns being handled too late\n- Large tasks producing broad, hard-to-review diffs\n- Inefficient token usage when the same model is applied to tasks of varying complexity\n\nThis repository provides a small orchestrator-based pattern that keeps the workflow structured while remaining easy to inspect, modify, and run.\n\n## What this provides\n\n- **A very simple orchestrator pattern** - A central agent coordinates planning, approval, implementation, and review phases that you can build upon and customize as needed.\n- **Scoped subagents** - Specialized agents handle planning, execution, testing, documentation, security, and code review\n- **Token-conscious delegation** - Subagents receive focused tasks instead of the full problem context whenever possible\n- **Approval-gated planning** - Non-trivial work is planned before implementation starts\n- **Minimal framework overhead** - The system is implemented as OpenCode configuration, agent prompts, skills, and a small plugin\n- **Reusable development skills** - Shared skill definitions for delegation, task management, Python quality, and security investigation\n\n## Design goals\n\nThis project is intentionally small. It is not trying to be a general-purpose agent platform.\n\nIt aims to:\n\n- Keep orchestration understandable\n- Reduce unnecessary context passed between agents\n- Preserve code quality through explicit planning and review steps\n- Make agent responsibilities clear\n- Avoid excessive automation that hides what is happening\n- Provide a useful starting point for customizing OpenCode workflows\n- Sync the improvements with my own personal workflow.\n\n## How the workflow works\n\nThe default entry point is the `orchestrator` agent.\n\nFor non-trivial tasks, the workflow typically follows this pattern:\n\n1. **Orchestrator receives the user request**\n2. **Planning is delegated** to a planning subagent\n3. **A concrete plan is written** under `.opencode\u002Fplans\u002F`\n4. **The user approves or rejects the plan**\n5. **Implementation is delegated** in scoped slices\n6. **Review agents validate the result** for correctness, tests, documentation impact, and security concerns\n\nThis keeps the main agent focused on coordination instead of forcing one large prompt\u002Fsession to handle every phase of the task.\n\n## Structure\n\n```text\n├── 📁 agents                                      # Agent definitions\n│   ├── 📝 api-docs-researcher.md\n│   ├── 📝 code-executor.md                        # Implementation subagent        \n│   ├── 📝 code-explorer.md                        # Read-only subagent\n│   ├── 📝 code-reviewer.md                         \n│   ├── 📝 docs-reviewer.md\n│   ├── 📝 host-security-investigator.md           \n│   ├── 📝 orchestrator.md                         # Workflow coordinator (No writting. Only analysis, context managment and delegation to subagents)\n│   ├── 📝 plan-runner.md                          # Plan drafting subagent apart from opencode Plan default agent, so we don't mess with that.\n│   ├── 📝 security-reviewer.md                    \n│   ├── 📝 spec-critic.md\n│   └── 📝 test-verifier.md\n├── 📁 command                                     # Optional OpenCode command hooks (may be empty)  \n├── 📁 plugins\n│   └── 📄 plan-post-approval.ts\n├── 📁 skills\n│   ├── 📁 agent-delegation\n│   │   └── 📝 SKILL.md\n│   ├── 📁 context7\n│   │   ├── 📝 README.md\n│   │   ├── 📝 SKILL.md\n│   │   ├── 📝 library-registry.md\n│   │   └── 📝 navigation.md\n│   ├── 📁 pythonic-quality\n│   │   └── 📝 SKILL.md\n│   ├── 📁 security-investigation\n│   │   ├── 📁 references\n│   │   │   └── 📝 vps-checklist.md\n│   │   ├── 📁 scripts\n│   │   │   └── 📄 vps-security-scan.sh\n│   │   └── 📝 SKILL.md\n│   ├── 📁 skill-creator\n│   │   ├── 📁 references\n│   │   │   ├── 📝 output-patterns.md\n│   │   │   └── 📝 workflows.md\n│   │   ├── 📁 scripts\n│   │   │   ├── 🐍 init_skill.py\n│   │   │   ├── 🐍 package_skill.py\n│   │   │   └── 🐍 quick_validate.py\n│   │   ├── 📄 LICENSE.txt\n│   │   └── 📝 SKILL.md\n│   └── 📁 task-management\n│       ├── 📁 scripts\n│       │   ├── 📄 migrate-schema.ts\n│       │   └── 📄 task-cli.ts\n│       ├── 📝 SKILL.md\n│       └── 📄 router.sh\n├── ⚙️ .gitignore\n├── 📝 AGENTS.md                                   # Global agent rules and delegation guidelines\n├── 📄 LICENSE\n├── 📝 README.md\n├── 📄 opencode.jsonc                              # Main OpenCode configuration\n├── ⚙️ package-lock.json\n├── ⚙️ package.json                                # Plugin \u002F tooling dependencies\n├── ⚙️ tsconfig.json                               # Typescript configuration          \n└── ⚙️ tui.json                                    # TUI settings (if used)\n```\n\n**Local-only skills (not tracked; see `.gitignore`):** add `skills\u002Fcontext7\u002F` or `skills\u002Fgitnexus-*` on your machine if you use those toolchains. Agent configs may still reference `gitnexus-*` in skill permissions.\n\n## Agents\n\n### Primary agents (Strong models) - Currently Using Kimi K2.6\u002FDeepSeek V4 Pro\u002FGLM 5.1\n\n- **orchestrator** - Coordinates multi-phase work through plan files, approval gates, and implementation slices\n- **build** - OpenCode Default Agent (Best for simple tasks)\n- **plan** - OpenCode Default Agent  (Best for simple tasks)\n\n### Subagents (Cheap models) - Currently Using DeepSeek V4 Flash\n\n- **plan-runner** - Drafts implementation plans under `.opencode\u002Fplans\u002F`\n- **code-executor** - Implements scoped coding tasks with minimal diffs\n- **test-verifier** - Validates changes through tests, linting, and type checking\n- **code-reviewer** - Reviews diffs for correctness, maintainability, and implementation risk\n- **docs-reviewer** - Checks whether documentation needs to be updated\n- **security-reviewer** - Identifies security risks in application code\n- **spec-critic** - Challenges plans before coding starts\n- **api-docs-researcher** - Researches external API documentation\n- **host-security-investigator** - Assesses hosting and infrastructure security concerns\n\n## When to use this\n\nThis configuration is useful when you want:\n\n- More structure than a single-agent coding workflow\n- Planning before implementation\n- Smaller, easier-to-review diffs\n- Separate review passes for tests, docs, code quality, and security\n- A multi-agent setup that is still simple enough to understand and modify\n\nIt may be unnecessary for very small edits where a direct `build` agent is faster and cheaper.\n\n## Plugins\n\n### Plan post-approval handoff\n\n`plugins\u002Fplan-post-approval.ts` automates the handoff after a plan is approved through `PlanApprove`.\n\nIt:\n\n- Extracts plan file paths from approval questions\n- Uses the last user message `agent` field as routing context\n- Hands off `plan` sessions to `build` after idle using `session.summarize` and `session.prompt`\n- Avoids duplicate Phase B automation for `orchestrator` sessions when `agent.orchestrator.plan_post_approval_handoff_agent` is set to `orchestrator`\n- Falls back to `plan_post_approval_handoff_agent` from `opencode.jsonc`, under `agent.orchestrator` or root configuration\n- Retries `session.prompt` with backoff\n\n## Skills\n\n- **agent-delegation** - Decision table for routing work to the appropriate subagent\n- **task-management** - CLI for tracking feature subtasks with dependencies\n- **pythonic-quality** - Pythonic idioms, SOLID design, and Liskov-safe patterns\n- **security-investigation** - Security audit orchestration\n- **skill-creator** - Guide for creating effective skills\n\n## Installation\n\n1. Install [OpenCode](https:\u002F\u002Fopencode.ai)\n\n2. Clone this repository to your OpenCode config directory:\n\n   ```bash\n   git clone \u003Crepo-url> ~\u002F.config\u002Fopencode\n   ```\n\n3. Install dependencies:\n\n   ```bash\n   cd ~\u002F.config\u002Fopencode\n   npm install\n   ```\n\n## Usage\n\nStart OpenCode and describe your task.\n\nThe `orchestrator` agent is configured as the default entry point. For simple tasks, the system can proceed directly. For larger or riskier tasks, it routes work through planning, approval, implementation, and review phases.\n\nThe intended flow is:\n\n```text\nuser request\n   ↓\norchestrator\n   ↓\nplan-runner \u002F plan\n   ↓\nuser approval\n   ↓\ncode-executor \u002F build\n   ↓\nreview subagents\n```\n\n## Configuration\n\nKey settings in `opencode.jsonc`:\n\n- **`default_agent: \"orchestrator\"`** - Sets the orchestrator as the default entry point\n- **`permission`** - Provides a minimal deny-by-default workspace baseline; each agent declares its tool policy in `agents\u002F\u003Cid>.md` frontmatter\n- **`agent.*.model`** - Configures model selection for primary agents and selected subagents\n- **`reasoningEffort`**, **`textVerbosity`**, and **`temperature`** - Tune agent behavior where needed\n- **`agent.orchestrator.plan_post_approval_handoff_agent`** - Controls post-approval routing for the plan handoff plugin\n\n\n## Changelog\n\n### [1.1.0] — 2026-05-12\n\n- **Changed**\n  - Orchestrator agent now don't have write permissions by default so we can force subagent usage.\n  - Updated agent configurations (orchestrator and all subagents)\n  - Updated `opencode.jsonc` with refined model routing, reasoning effort, and permission settings\n  - Updated README with structure, workflow, and configuration documentation\n- **Fixed**\n  - Corrected model names in agent configurations\n  - Added DeepSeek max reasoning effort support in `opencode.jsonc`\n- **Removed**\n  - Build and Plan customization. The config now uses the opencode default Build and Plan agents. Add `agents\u002Fbuild.md` and `agents\u002Fplan.md` if you want to override.\n\n### [1.0.0] — 2026-05-02\n\nInitial release: multi-agent orchestrator pattern with planning, code execution, and review subagents.\n\n","这是一个轻量级的多代理工作流编排实现，通过中央协调器将任务分配给专门的子代理，以优化令牌效率而不牺牲代码生成质量。项目采用TypeScript编写，核心功能包括一个简单的编排模式、专注于特定任务的子代理以及令牌意识的任务委派。它适用于需要规划、执行、审查和安全检查等多阶段协作的软件开发场景，特别是当单一代理处理复杂或模糊任务变得不可靠时。该项目旨在保持工作流程结构化的同时易于检查、修改和运行，适合希望在不增加过多框架开销的情况下提高AI辅助开发效率的开发者使用。","2026-06-11 04:05:24","CREATED_QUERY"]