by NicoGenti
Cost-aware multi-agent orchestrator kit for OpenCode CLI — one routing agent, 14 specialized subagents, zero application code touched by the router.
Available after the catalog security scan passes.
⚠️ Third-Party Software Notice
This skill is third-party open-source software developed and hosted independently on GitHub. SkillsLLM is an informational directory and does not control or maintain the underlying repository.
Any security checks, ratings, or warnings displayed by SkillsLLM are automated and limited in scope. They do not constitute a security certification or guarantee that the software is safe, error-free, or free from malicious code, vulnerabilities, compromised dependencies, or prompt-injection risks.
Review the source code, permissions, dependencies, and configuration before installing or running any third-party skill. Use is at your own risk. To the maximum extent permitted by applicable law, SkillsLLM is not liable for losses arising from third-party software.
The deep catalog scan for this skill is still queued. Run an instant dependency check now instead.
# Add to your Claude Code skills
git clone https://github.com/NicoGenti/opencode-orchestrator-kitGuides for using ai agents skills like opencode-orchestrator-kit.
See how opencode-orchestrator-kit compares with popular alternatives.
A cost-aware, multi-agent workflow for OpenCode: one router agent that never writes code, and specialists that each run on the cheapest model able to do the job.
Most "AI does everything" setups burn tokens: one powerful model handles research, planning, coding and review inside a single long-lived context. Quality drifts as the context grows, and every step is billed at the most expensive rate.
The kit splits the work into a routing layer and a roster of specialists:
orchestrator-quick (reach it with /quick) applies the same discipline without the plan board and the session memory.profiler generates a compact .context/repo-map.md (hard 300-line budget) that the explorer reads before broad searches, reducing whole-file reads and token waste.profiler agent detects the stack (or interviews you for an empty repo) and scaffolds .context/ (memory) and plan/ (kanban).On OpenCode 2 the kit is the opencode-orchestrator-kit plugin, published on npm: agents, commands and skills are registered at runtime — nothing is written into your project's .opencode/ directory. Add "package": "opencode-orchestrator-kit" to the plugins array (below). A local plugin/ directory and a git install remain available as alternatives. The native copy-based installer still works and remains supported, but it is deprecated in favor of the plugin on OpenCode 2 (the default profile it ships references provider IDs that are not yours — see the Choose your model profile step). Optional notes for OpenCode Studio users are in docs/SETUP-OPENCODE-STUDIO.md.
"Add refresh-token rotation to the auth service."
| Step | Agent | What happens |
|---|---|---|
| 1 | explorer |
Reads the current auth flow (read-only). |
| 2 | planner |
Turns findings into a phased plan in plan/draft/. |
| 3 | plan-writer |
Owns the plan board: plan/draft/ -> plan/in-progress/ -> plan/qa/ -> plan/complete/. |
| 4 | developer-fixer |
Implements one phase at a time, with fresh context each time. |
| 5 | test-engineer |
Writes and runs tests for the new logic. |
| 6 | security |
Reviews the auth-sensitive change before it reaches plan/complete/. |
| 7 | comprehension-coach |
Optional human comprehension check, scaled to the change (NONE / LIGHT / DEEP). |
Every delegation uses a 9-section task spec (Goal, Success Criteria, Scope, Safety, Inputs, Outputs, Test Plan, Verification, Edge Cases), never a free-form prompt. The orchestrator itself never edits auth.ts.
agents/orchestrator.md is the default mode: primary agent and can only classify and delegate, so orchestration mistakes cannot silently change the codebase. agents/orchestrator-quick.md is the opt-in alternate for small tasks, with the same write surface and a shorter cycle.{{TIER_ROUTER}}, {{TIER_REASONING}}, {{TIER_CODE}}, {{TIER_FAST}} and {{TIER_REVIEW}}. Switching provider means editing one config file, and a validator runs before install.plan/ is the source of truth, so work is resumable and reviewable in Git.code-reviewer and security verify the code; comprehension-coach checks that the developer understood it.| Concern | Primary agent | Notes |
|---|---|---|
| Orchestration (routing, checkpointing) | orchestrator |
Default entry point. Never writes application code. |
| Orchestration, small tasks (one delegation) | orchestrator-quick |
Opt-in via /quick. Same write surface, no plan board. |
| Exploration (codebase, file, symbol research) | explorer |
Read-only. |
| Planning (turns exploration into phased plans) | planner |
Writes to plan/draft/ only. |
| Plan board (create, move, check, adopt plan files) | plan-writer |
Sole owner of the board. |
| Implementation (one phase at a time, TDD) | developer-fixer |
Follows the 9-section task spec. |
| Testing (test authoring, coverage, reproduction) | test-engineer |
|
| Review (general correctness and quality) | code-reviewer |
Distinct from security review. |
| Security (vulnerabilities, threat model, hardening) | security |
Routed for auth, crypto, input handling and secrets changes. |
| Human comprehension (post-review check) | comprehension-coach |
Read-only. Distinct from technical verification. |
Full contracts, permissions and model fallbacks are in each agents/*.md file (the pc-doctor and writer specialists live in extras/*.md) and are summarized in AGENTS.md.
| Tier | Agents | Installed by default? |
|---|---|---|
| Core routing | orchestrator, orchestrator-quick, profiler, explorer, oracle, planner |
Yes |
| Core delivery | developer-fixer, test-engineer, code-reviewer, security, comprehension-coach, plan-writer |
Yes |
| Operations helpers | build-helper, npm-helper, deploy-helper |
Yes (routed only on a matching failure) |
| Docs lookup | librarian |
Yes (routed only for remote documentation lookups) |
| Opt-in extras | pc-doctor (Windows-local only), writer (docs generation) |
--with-extras only |
With the plugin, tier resolution happens at runtime from plugin/src/tiers.ts presets — there is no {{TIER_*}} substitution at install time.
opencode-orchestrator-kit is published on npm (1.1.1, tag v1.1.1; later releases follow .github/workflows/release.yml and the steps in CONTRIBUTING.md). OpenCode 2 loads plugins from the plugins array in opencode.json or opencode.jsonc (Plugins). A string entry is a package name or a local directory. Pass options with the object form { "package", "options" }. Local paths must be directories (a path to a single file is skipped). Relative paths resolve from the config file that contains the entry.
Primary install, the npm package:
{
"$schema": "https://opencode.ai/config.json",
"plugins": [
{
"package": "opencode-orchestrator-kit",
"options": { "preset": "generic", "withExtras": false },
},
],
}
A local checkout of this repository uses the same object form with an absolute directory ("package": "/absolute/path/to/opencode-orchestrator-kit/plugin"). The OpenCode 2 docs also accept a git package spec on opencode plugin add, including a subdirectory selector, because this plugin lives in plugin/ and not at the repository root:
opencode plugin add 'github:NicoGenti/opencode-orchestrator-kit#main::path:plugin'
That command writes a package entry. Set preset and withExtras with the object form ab
opencode-orchestrator-kit is an open-source ai agents skill for AI coding assistants such as Claude Code, Codex CLI, and ChatGPT, built by NicoGenti. Cost-aware multi-agent orchestrator kit for OpenCode CLI — one routing agent, 14 specialized subagents, zero application code touched by the router. It has 9 GitHub stars.
opencode-orchestrator-kit's catalog security scan is still queued. You can run an instant dependency and prompt-injection check now with the "Scan for vulnerabilities" button above.
Clone the repository with "git clone https://github.com/NicoGenti/opencode-orchestrator-kit" and add it to your Claude Code skills directory (see the Installation section above).
opencode-orchestrator-kit is primarily written in TypeScript. It is open-source under NicoGenti on GitHub, so you can review or fork the full source.
Yes. SkillsLLM lists many other AI Agents skills you can browse and compare side by side. Open the AI Agents category from the badge at the top of this page, or use the Related Skills and comparison links further down to weigh opencode-orchestrator-kit against similar tools.
No comments yet. Be the first to share your thoughts!