orchestrate

An agent skill by cursor, from cursor/plugins. Tags: ai, automation, multi-agent, planning, workflow.

What it does

Use only when the user explicitly types `/orchestrate <goal>` to decompose a large task, spawn a tree of parallel cloud-agent workers/subplanners/verifiers via the Cursor SDK, and collect structured handoffs; do not invoke autonomously.

Install

With the skills CLI, which installs into Claude Code, Codex, Cursor and other agents:

npx skills add cursor/plugins --skill orchestrate

Or copy the skill folder into Claude Code's skills directory by hand (~/.claude/skills for every project, or .claude/skills inside one):

git clone --depth 1 https://github.com/cursor/plugins
cp -r plugins/orchestrate/skills/orchestrate ~/.claude/skills/orchestrate

Safety box score

Grade D, elevated risk: 65 of 100 risk points, rated 2026-10-16. The grade rates what the skill and its scripts can reach on the machine of whoever installs it, across eight categories from shell execution to secrets access. How the score works.

This skill orchestrates a tree of cloud Cursor agents, persists workspace state/handoffs, and mirrors activity into Slack. The main risks are broad remote autonomy, account-scoped access via `CURSOR_API_KEY`/`SLACK_BOT_TOKEN`, and the fact that it re-injects goals and handoffs into agent prompts.

Category levels, from 0 (none) to 3 (high)
Category	Level	Why
Code & shell execution	2 (moderate)	The CLI and helpers run native commands such as `execFileSync("git", ...)` and `bun`-driven scripts, but the visible code does not execute arbitrary downloaded code locally.
Network access	2 (moderate)	It talks to Cursor and Slack (`Agent.create`, `new WebClient(token)`) and fetches from origin with `git fetch`, so task text, handoffs, and status leave the machine over the network.
Filesystem reach	2 (moderate)	It writes orchestration state and artifacts (`plan.json`, `state.json`, `handoffs/*.md`, generated schemas), with some user-level home-state behavior hinted by the operator-mode tests.
Secrets & credentials	2 (moderate)	Setup requires a personal `CURSOR_API_KEY` and optional `SLACK_BOT_TOKEN`, both read directly from `process.env` and used for remote API access.
Privilege & persistence	1 (low)	No sudo, services, or global installs are visible; the main non-workspace state is user-level configuration such as operator mode and git-based operations.
Instruction hijack surface	2 (moderate)	The prompt files feed goals, handoffs, and Slack context back into spawned agents, and some instructions explicitly push autonomy (for example `Default to autonomous` and `probe, ignore`).
Irreversible actions	3 (high)	The skill’s purpose is to `spawn a tree of parallel cloud-agent workers/subplanners/verifiers`, and `kill-tree` can cancel every running agent in that tree.
Transparency	2 (moderate)	Several core orchestration files were not readable in the provided sample (`scripts/core/agent-manager.ts`, `scripts/cli/task.ts`, `scripts/cli/util.ts`, `scripts/core/loop.ts`, etc.), so some behavior could not be fully audited.

Before installing

Read `SKILL.md`, `scripts/cli/forensics.ts`, and `scripts/core/andon.ts` to confirm you accept the agent-cancellation and git-fetch behavior.
Use a throwaway Cursor/Slack token first: set `CURSOR_API_KEY` and `SLACK_BOT_TOKEN` to test values and run `bun cli.ts models` / `bun cli.ts tree <workspace>` on a scratch workspace.
Inspect the prompt templates in `prompts/*.md` so you understand how your goal, handoffs, and Slack context are fed back into spawned agents.
If you rely on the home-directory operator flag, check where it lives and whether you’re comfortable with user-level state outside the workspace.

Rated 2026-10-16 by an automated review of 57 files from the skill's repository.

Source

Repository: cursor/plugins (all skills from this repository)
Path: orchestrate/skills/orchestrate/SKILL.md
Branch: main
Collection: cursor
Updated: 2026-09-23

ralph-loop — Start a Ralph Loop for iterative self-referential development.
continual-learning — Orchestrate continual learning by delegating transcript mining and AGENTS.md updates to `agents-memory-updater`.
swarm — Fan out N parallel workers, drain them, and return one report. Use for /swarm, 'swarm this', or parallel coverage, races, gauntlets, and exploration.
agents-build — Use to extend an existing agent project with memory, app integration, VPC, multi-agent, migration, model, browser, code interpreter, payments.
marketing-loops — When the user wants to set up a recurring, self-running marketing workflow — a repeatable loop an AI agent runs on a cadence (weekly, daily.
autoplan — Auto-review pipeline — reads the full CEO, design, eng.