
openai-symphony-autonomous-agents
Symphony turns project work into isolated, autonomous implementation runs, allowing teams to manage work instead of supervising coding agents.
- Installs
- 1.4K
- Last 24h
- —
- Trending rank
- —
- Security
- Not audited
Install
Works with every agent the skills CLI supports.
npx skills add https://github.com/reason-machines/trending-skills --skill openai-symphony-autonomous-agents- 01
Install
Run the command in your project. The CLI asks which agents to install for.
- 02
Describe the task
Ask your agent for something this skill covers. It loads openai-symphony-autonomous-agents automatically when relevant.
- 03
Review the result
The agent follows the SKILL.md steps, scripts and references bundled with the skill.
SKILL.md
The instructions your agent reads when the skill is triggered.
OpenAI Symphony
Skill by ara.so — Daily 2026 Skills collection.
Symphony turns project work into isolated, autonomous implementation runs, allowing teams to manage work instead of supervising coding agents. Instead of watching an agent code, you define tasks (e.g. in Linear), and Symphony spawns agents that complete them, provide proof of work (CI status, PR reviews, walkthrough videos), and land PRs autonomously.
What Symphony Does
- Monitors a work tracker (e.g. Linear) for tasks
- Spawns isolated agent runs per task (using Codex or similar)
- Each agent implements the task, opens a PR, and provides proof of work
- Engineers review outcomes, not agent sessions
- Works best in codebases using harness engineering
Installation Options
Option 1: Ask an agent to build it
Paste this prompt into Claude Code, Cursor, or Codex:
Implement Symphony according to the following spec:
https://github.com/openai/symphony/blob/main/SPEC.md
Option 2: Use the Elixir reference implementation
git clone https://github.com/openai/symphony.git
cd symphony/elixir
Follow elixir/README.md, or ask an agent:
Set up Symphony for my repository based on
https://github.com/openai/symphony/blob/main/elixir/README.md
Elixir Reference Implementation Setup
Requirements
- Elixir + Mix installed
- An OpenAI API key (for Codex agent)
- A Linear API key (if using Linear integration)
- A GitHub token (for PR operations)
Environment Variables
export OPENAI_API_KEY="sk-..." # OpenAI API key for Codex
export LINEAR_API_KEY="lin_api_..." # Linear integration
export GITHUB_TOKEN="ghp_..." # GitHub PR operations
export SYMPHONY_REPO_PATH="/path/to/repo" # Target repository
Install Dependencies
cd elixir
mix deps.get
Configuration (elixir/config/config.exs)
import Config
config :symphony,
openai_api_key: System.get_env("OPENAI_API_KEY"),
linear_api_key: System.get_env("LINEAR_API_KEY"),
github_token: System.get_env("GITHUB_TOKEN"),
repo_path: System.get_env("SYMPHONY_REPO_PATH", "./"),
poll_interval_ms: 30_000,
max_concurrent_agents: 3
Run Symphony
mix symphony.start
# or in IEx for development
iex -S mix
Core Concepts
Isolated Implementation Runs
Each task gets its own isolated run:
- Fresh git branch per task
- Agent operates only within that branch
- No shared state between runs
- Proof of work collected before PR merge
Proof of Work
Before a PR is accepted, Symphony collects:
- CI/CD pipeline status
- PR review feedback
- Complexity analysis
- (optionally) walkthrough videos
Key Elixir Modules & Patterns
Starting the Symphony supervisor
# In your application.ex or directly
defmodule MyApp.Application do
use Application
def start(_type, _args) do
children = [
Symphony.Supervisor
]
Supervisor.start_link(children, strategy: :one_for_one)
end
end
Defining a Task (Symphony Task struct)
defmodule Symphony.Task do
@type t :: %__MODULE__{
id: String.t(),
title: String.t(),
description: String.t(),
source: :linear | :manual,
status: :pending | :running | :completed | :failed,
branch: String.t() | nil,
pr_url: String.t() | nil,
proof_of_work: map() | nil
}
defstruct [:id, :title, :description, :source,
status: :pending, branch: nil,
pr_url: nil, proof_of_work: nil]
end
Spawning an Agent Run
defmodule Symphony.AgentRunner do
@doc """
Spawns an isolated agent run for a given task.
Each run gets its own branch and Codex session.
"""
def run(task) do
branch = "symphony/#{task.id}-#{slugify(task.title)}"
with :ok <- Git.create_branch(branch),
{:ok, result} <- Codex.implement(task, branch),
{:ok, pr_url} <- GitHub.open_pr(branch, task),
{:ok, proof} <- ProofOfWork.collect(pr_url) do
{:ok, %{task | status: :completed, pr_url: pr_url, proof_of_work: proof}}
else
{:error, reason} -> {:error, reason}
end
end
defp slugify(title) do
title
|> String.downcase()
|> String.replace(~r/[^a-z0-9]+/, "-")
|> String.trim("-")
end
end
Linear Integration — Polling for Tasks
defmodule Symphony.Linear.Poller do
use GenServer
@poll_interval Application.compile_env(:symphony, :poll_interval_ms, 30_000)
def start_link(opts \\ []) do
GenServer.start_link(__MODULE__, opts, name: __MODULE__)
end
def init(_opts) do
schedule_poll()
{:ok, %{processed_ids: MapSet.new()}}
end
def handle_info(:poll, state) do
case Symphony.Linear.Client.fetch_todo_tasks() do
{:ok, tasks} ->
new_tasks = Enum.reject(tasks, &MapSet.member?(state.processed_ids, &1.id))
Enum.each(new_tasks, &Symphony.AgentRunner.run/1)
new_ids = Enum.reduce(new_tasks, state.processed_ids, &MapSet.put(&2, &1.id))
schedule_poll()
{:noreply, %{state | processed_ids: new_ids}}
{:error, reason} ->
Logger.error("Linear poll failed: #{inspect(reason)}")
schedule_poll()
{:noreply, state}
end
end
defp schedule_poll do
Process.send_after(self(), :poll, @poll_interval)
end
end
Linear API Client
defmodule Symphony.Linear.Client do
@linear_api "https://api.linear.app/graphql"
def fetch_todo_tasks do
query = """
query {
issues(filter: { state: { name: { eq: "Todo" } } }) {
nodes {
id
title
description
}
}
}
"""
case HTTPoison.post(@linear_api, Jason.encode!(%{query: query}), headers()) do
{:ok, %{status_code: 200, body: body}} ->
tasks =
body
|> Jason.decode!()
|> get_in(["data", "issues", "nodes"])
|> Enum.map(&to_task/1)
{:ok, tasks}
{:error, reason} ->
{:error, reason}
end
end
defp headers do
[
{"Authorization", System.get_env("LINEAR_API_KEY")},
{"Content-Type", "application/json"}
]
end
defp to_task(%{"id" => id, "title" => title, "description" => desc}) do
%Symphony.Task{id: id, title: title, description: desc, source: :linear}
end
end
Proof of Work Collection
defmodule Symphony.ProofOfWork do
@doc """
Collects proof of work for a PR before it can be merged.
Returns a map with CI status, review feedback, and complexity.
"""
def collect(pr_url) do
with {:ok, ci_status} <- wait_for_ci(pr_url),
{:ok, reviews} <- fetch_reviews(pr_url),
{:ok, complexity} <- analyze_complexity(pr_url) do
{:ok, %{
ci_status: ci_status,
reviews: reviews,
complexity: complexity,
collected_at: DateTime.utc_now()
}}
end
end
defp wait_for_ci(pr_url, retries \\ 30) do
case GitHub.get_pr_ci_status(pr_url) do
{:ok, :success} -> {:ok, :success}
{:ok, :pending} when retries > 0 ->
Process.sleep(60_000)
wait_for_ci(pr_url, retries - 1)
{:ok, status} -> {:ok, status}
{:error, reason} -> {:error, reason}
end
end
defp fetch_reviews(pr_url), do: GitHub.get_pr_reviews(pr_url)
defp analyze_complexity(pr_url), do: GitHub.get_pr_diff_complexity(pr_url)
end
Implementing the SPEC.md (Custom Implementation)
When building Symphony in another language, the spec defines:
- Task Source — poll Linear/GitHub/Jira for tasks in a specific state
- Agent Invocation — call Codex (or another agent) with task context
- Isolation — each run on a fresh branch, containerized if possible
- Proof of Work — CI, review, and analysis before merge
- Landing — auto-merge or present to engineer for approval
Minimal implementation loop in pseudocode:
# Core symphony loop
def symphony_loop(state) do
tasks = fetch_new_tasks(state.source)
tasks
|> Enum.filter(&(&1.status == :todo))
|> Enum.each(fn task ->
Task.async(fn ->
branch = create_isolated_branch(task)
invoke_agent(task, branch) # Codex / Claude / etc.
proof = collect_proof_of_work(branch)
present_for_review(task, proof)
end)
end)
Process.sleep(state.poll_interval)
symphony_loop(state)
end
Common Patterns
Limiting Concurrent Agent Runs
defmodule Symphony.AgentPool do
use GenServer
@max_concurrent 3
def start_link(_), do: GenServer.start_link(__MODULE__, %{running: 0, queue: []}, name: __MODULE__)
def submit(task) do
GenServer.cast(__MODULE__, {:submit, task})
end
def handle_cast({:submit, task}, %{running: n} = state) when n < @max_concurrent do
spawn_agent(task)
{:noreply, %{state | running: n + 1}}
end
def handle_cast({:submit, task}, %{queue: q} = state) do
{:noreply, %{state | queue: q ++ [task]}}
end
def handle_info({:agent_done, _result}, %{running: n, queue: [next | rest]} = state) do
spawn_agent(next)
{:noreply, %{state | running: n, queue: rest}}
end
def handle_info({:agent_done, _result}, %{running: n} = state) do
{:noreply, %{state | running: n - 1}}
end
defp spawn_agent(task) do
parent = self()
spawn(fn ->
result = Symphony.AgentRunner.run(task)
send(parent, {:agent_done, result})
end)
end
end
Manual Task Injection (No Linear)
# In IEx or a Mix task
Symphony.AgentPool.submit(%Symphony.Task{
id: "manual-001",
title: "Add rate limiting to API",
description: "Implement token bucket rate limiting on /api/v1 endpoints",
source: :manual
})
Troubleshooting
| Problem | Likely Cause | Fix |
|---|---|---|
| Agents not spawning | Missing OPENAI_API_KEY | Check env var is exported |
| Linear tasks not detected | Wrong Linear state filter | Update query filter to match your board's state name |
| PRs not opening | Missing GITHUB_TOKEN or wrong repo | Verify token has repo scope |
| CI never completes | Timeout too short | Increase retries in wait_for_ci/2 |
| Too many concurrent runs | Default pool size | Set max_concurrent_agents in config |
| Branch conflicts | Agent reusing branch names | Ensure task IDs are unique per run |
Debug Mode
# In config/dev.exs
config :symphony, log_level: :debug
# Or at runtime
Logger.put_module_level(Symphony.AgentRunner, :debug)
Logger.put_module_level(Symphony.Linear.Poller, :debug)
Resources
- SPEC.md — Full Symphony specification
- elixir/README.md — Elixir setup guide
- Harness Engineering — Prerequisite methodology
- Apache 2.0 License
FAQ
What does the openai-symphony-autonomous-agents skill do?
Symphony turns project work into isolated, autonomous implementation runs, allowing teams to manage work instead of supervising coding agents.
How do I install openai-symphony-autonomous-agents?
Run npx skills add https://github.com/reason-machines/trending-skills --skill openai-symphony-autonomous-agents in your project. The skills CLI detects the agents you use and copies the skill into the right folder. For Claude Code you can also place the skill folder at ~/.claude/skills/openai-symphony-autonomous-agents/ (personal) or .claude/skills/openai-symphony-autonomous-agents/ (project).
Which AI agents can use openai-symphony-autonomous-agents?
Any agent that supports the open Agent Skills format (a folder with a SKILL.md file), including Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, Windsurf, Cline, OpenCode and more.
Is openai-symphony-autonomous-agents safe to install?
No partner security audit is available yet. Skills can run commands and scripts, so read the SKILL.md and any bundled files before installing.
Who maintains openai-symphony-autonomous-agents?
It is published by reason-machines in reason-machines/trending-skills. It has 1,406 recorded installs.
More from reason-machines/trending-skills
openclaw-control-center
reason-machines/trending-skills
Local-first, security-first control center for OpenClaw agents — visibility dashboard with readonly defaults, token attribution, collaboration tracing, and safe write operations.
inkos-multi-agent-novel-writing
reason-machines/trending-skills
Multi-agent CLI system for autonomous novel writing, auditing, and revision with human review gates
claude-hud-statusline
reason-machines/trending-skills
A Claude Code plugin that displays a real-time HUD showing context usage, active tools, running agents, and todo progress in your terminal statusline.
understand-anything-knowledge-graph
reason-machines/trending-skills
Turn any codebase into an interactive knowledge graph using Claude Code skills — explore, search, and ask questions about any project visually.
agency-agents-ai-specialists
reason-machines/trending-skills
A collection of specialized AI agent personalities for Claude Code, Cursor, Aider, Windsurf, and other AI coding tools — covering engineering, design, marketing, sales, and more.
agent-browser-automation
reason-machines/trending-skills
Headless browser automation CLI for AI agents using native Rust binary with Chrome DevTools Protocol
Popular ai & agents skills
google-agents-cli-publish
google/agents-cli
This skill should be used when the user wants to "publish an agent", "publish my ADK agent", "register an agent with Gemini", "publish to Gemini" (formerly Gemini Enterprise), or needs guidance on the agents-cli publish gemini command. Also use when the user wants to "manage agents in Agent Registry", "list/update/delete registered agents", or "register an MCP server". Covers ADK vs A2A registration modes, programmatic and interactive usage, flag reference, auto-detection from deployment metadata, Agent Registry fleet management, and troubleshooting. Part of the agents-cli skills suite. Do NOT use for deployment (use google-agents-cli-deploy).
google-agents-cli-eval
google/agents-cli
This skill should be used when the user wants to "run an evaluation", "evaluate my agent", "evaluate my ADK agent", "write an eval dataset", "analyze eval failures", "compare eval results", "optimize agent", or needs guidance on the Gemini platform eval methodology and the Quality Flywheel. Covers eval metrics, dataset schema, LLM-as-judge scoring, and common failure causes. Applies to any agents-cli project, whatever framework the agent is written in. Do NOT use for agent API code patterns (ADK: use google-agents-cli-adk-code), deployment (use google-agents-cli-deploy), or project scaffolding (use google-agents-cli-scaffold).
skill-creator
anthropics/skills
Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.
codex-pet
prime-skills/runcomfy-agent-skills
Codex Pet generator on RunComfy. Build a Codex-compatible Codex Pet spritesheet.webp + pet.json from a single reference image, drop it into ${CODEXHOME:-$HOME/.codex}/pets/ / and Codex picks it up as a custom Codex Pet next to the 8 built-ins. This skill produces the exact Codex Pet atlas Codex expects (1536x1872 PNG/WebP, 8 cols x 9 rows, 192x208 cells, 9 animation states — idle, running-right, running-left, waving, jumping, failed, waiting, running, review). Calls OpenAI GPT Image 2 edit ONCE via the local RunComfy CLI as runcomfy run openai/gpt-image-2/edit to produce a canonical Codex Pet pose, then assembles all 9 animation rows programmatically with ImageMagick micro-transforms — no Codex Pro, no $imagegen, no OPENAIAPIKEY required, only RUNCOMFYTOKEN. Triggers on "codex pet", "create codex pet", "make codex pet", "hatch codex pet", "/hatch image", "desktop pet codex", "codex pets", "spritesheet.webp", or any explicit ask to build a custom pet for OpenAI Codex.
full-output-enforcement
leonxlnx/taste-skill
Overrides default LLM truncation behavior. Enforces complete code generation, bans placeholder patterns, and handles token-limit splits cleanly. Apply to any task requiring exhaustive, unabridged output.
executing-plans
obra/superpowers
Use when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available