---
title: "orrery"
description: "Configurable cross-client architect orchestration with lazy setup, exact user-selected native roles, risk-routed Astra and Terra implementation lanes, content-addressed tool-surface consent, transactional adapter install"
canonical: https://agentpluginsdirectory.com/plugins/orrery
last-updated: 2026-09-26
---

# orrery
Configurable cross-client architect orchestration with lazy setup, exact user-selected native roles, risk-routed Astra and Terra implementation lanes, content-addressed tool-surface consent, transactional adapter installation, and a separate explicit Codex Luna app-task lane.
- Slug: orrery
- Publisher: Divyam Talwar
- Repository: https://github.com/DivyamTalwar/ORRERY
- Manifest: plugins/orrery/plugin.json
- Version: 0.7.0
- License: MIT
- Category (editorial): agent-tooling
- Skills: 2 (orchestration, setup)
- MCP servers: 1 (orrery)
- Stars: 9
- Repository created: 2026-08-09
- Repository last pushed: 2026-09-05
- Publisher type: User
- Listing: https://agentpluginsdirectory.com/plugins/orrery
- Schema: https://agent-plugins.org/schemas/1.0.0/plugin.schema.json

## What orrery does, in the publisher's words

Architect-first orchestration for coding agents. Exact model pinning. Consented writes. Fail-closed, always.

What it guarantees: The one rule · Security model · Tool-surface consent · Supported clients · What it will not do · Interrogate it yourself · When not to use this

Orrery keeps you the architect.

- Every client defines subagents differently. One wants.codex/agents/*.toml, another.cursor/agents/*.md, another.claude/agents/*.md. They bind different things: some pin a model and reasoning effort and a sandbox, some pin only a model.
- A plugin that writes into your agent config is privileged. If a prompt-injected model can talk it into writing an arbitrary file to an arbitrary path, you have handed an attacker your editor.
- Astra ⊕ Terra: mutually exclusive. Exactly one implementer per delegation.
- Native lane ⊕ Luna lane: mutually exclusive. Luna is opt-in only and is never a fallback.

From the project README, punctuation lightly normalized. Full text: https://raw.githubusercontent.com/DivyamTalwar/ORRERY/HEAD/README.md

## Skills

- orchestration: Configurable cross-client architect workflow that inherits the parent model, loads saved exact native role preferences, routes routine versus high-complexity implementation, preserves the separate opt-in Luna app-task lane, and keeps parent verification and acceptance.
- setup: Run Orrery's first-use or reconfiguration interview in the parent chat, validate exact client-native model choices, persist logical preferences, preview native adapter files, and install only after explicit confirmation.

Descriptions come from the frontmatter of each SKILL.md, punctuation lightly normalized.

## MCP servers

- orrery: transport: stdio; command: bun ${PLUGIN_ROOT}/mcp/server.ts

Read from the plugin's own mcp.json. Environment variable names only, never values.