Data

Tools

Indexes

Applications

Experiments

Awesome

An open API service indexing awesome lists of open source software.

https://github.com/dubsopenhub/agent-orchestra

Multi-agent fleet conductor for the GitHub Copilot CLI.
https://github.com/dubsopenhub/agent-orchestra

Last synced: 2 days ago
JSON representation

Multi-agent fleet conductor for the GitHub Copilot CLI.

Awesome Lists containing this project

README 

          
**🎼 Agent Orchestra**: Multi-agent fleet conductor for the GitHub Copilot CLI. 



  🌐 Open the live Agent Orchestra site →




**Multi-agent fleet conductor for the GitHub Copilot CLI. 



Launch multiple visible Terminal Stampede commander groups, let them collaborate in real time, score output quality with sealed [Shadow Score](https://github.com/DUBSOpenHub/shadow-score-spec/blob/main/SPEC.md) gates, and judge run repeatability with [Fleet Scorecard](https://github.com/DUBSOpenHub/fleet-scorecard-spec/blob/main/SPEC.md).**



[![GitHub Copilot CLI](https://img.shields.io/badge/platform-Copilot%20CLI-232F3E.svg)](https://docs.github.com/copilot/concepts/agents/about-copilot-cli)

[![Live Site](https://img.shields.io/badge/live%20site-Agent%20Orchestra-7dd3fc?logo=githubpages&logoColor=white)](https://dubsopenhub.github.io/agent-orchestra/)

[![License: MIT](https://img.shields.io/badge/License-MIT-green.svg)](LICENSE)

[![Security Policy](https://img.shields.io/badge/Security-Policy-brightgreen?logo=github)](SECURITY.md)



> ### ⚡ One Command. That's It.

>

> **Never used the CLI before? No problem.** Follow these 3 steps:

>

> **1. Open your terminal**

> - 🍎 **Mac:** Press `⌘ + Space`, type **Terminal**, hit Enter

> - 🪟 **Windows:** Open **Git Bash** or **WSL** (the installer uses `bash`)

> - 🐧 **Linux:** Press `Ctrl + Alt + T`

>

> **2. Paste this line and press Enter:**

> ```bash

> curl -fsSL https://raw.githubusercontent.com/DUBSOpenHub/agent-orchestra/main/quickstart.sh | bash

> ```

> *Already have the CLI? No worries — this detects what is present and installs the local Agent Orchestra helpers.*

>

> **3. Watch the fleet:** `agent-orchestra-pulse`

>

> That's it — raise the baton. 🎼

>

> *Requires an active [Copilot subscription](https://github.com/features/copilot/plans) for live Copilot CLI fleet runs.*



---



## 🚀 30-Second Overview



Agent Orchestra is for work that is too big, risky, or cross-cutting for one agent thread:



- **Need multiple strategies at once?** It launches commander-led groups in visible terminal panes.

- **Need live collaboration, not isolated answers?** Commanders propose, review, improve, converge, and broadcast learnings through append-only ledgers.

- **Need real observability?** Agent Pulse and the Stampede monitor read concise run telemetry as the work happens.

- **Need quality gates?** [Shadow Score](https://github.com/DUBSOpenHub/shadow-score-spec/blob/main/SPEC.md) criteria are sealed before launch and applied only after commander bundles finish.

- **Need a run decision?** [Fleet Scorecard](https://github.com/DUBSOpenHub/fleet-scorecard-spec/blob/main/SPEC.md) answers what changed, what won, what failed, and whether to run it again.

- **Need honest partials?** A partial launch stays partial; Agent Orchestra never silently upgrades incomplete work to success.



No server. No message broker. No dashboard daemon. Just Copilot CLI skills, Terminal Stampede, tmux, and filesystem IPC.



---



## 📡 Watch Every Run Live with Agent Pulse



Agent Orchestra is best experienced with [**Agent Pulse**](https://github.com/DUBSOpenHub/copilot-cli-agent-pulse), the real-time terminal dashboard for Copilot CLI sessions and swarm runs.





  

    Agent Pulse dashboard showing live Copilot CLI sessions, agents, and swarm telemetry

  




Agent Pulse makes Agent Orchestra observable while it runs:



| What you see live | Why it matters |

|---|---|

| Commander status | Know which commander groups are running, partial, failed, or complete |

| Sub-agent counts | Track live, completed, failed, and seen sub-agents without ledger spelunking |

| Terminal Stampede runs | See active `.stampede/run-*` work across repos |

| Telemetry confidence | Distinguish authoritative live counts from recent/historical events |

| Recent launches | Spot bursts, stuck runs, and platform-cap pressure |



Install it alongside Agent Orchestra:



```bash

curl -fsSL https://raw.githubusercontent.com/DUBSOpenHub/copilot-cli-agent-pulse/main/quickstart.sh | bash

agentpulse

```



---



## 🐝 Dogfooded at Swarm Scale



Agent Orchestra follows the same swarm-scale operating model that shaped Agent Conductor: visible commander groups, bounded sub-agent fan-out, collaboration ledgers, live dashboard telemetry, Shadow Score sealing, Fleet Scorecard run decisions, install flow checks, and repo-readiness validation.



Those signals come from local Stampede run artifacts and compatibility ledgers, so the system can be tested without adding a server, queue, or hosted control plane.



---



## 🤔 What Is This?



Agent Orchestra coordinates multiple commander-led agent groups against the same mission. Each commander gets its own namespace, bounded sub-agent hierarchy, collaboration bus access, and final bundle contract. The orchestrator then synthesizes the best findings from commander bundles, collaboration ledgers, live telemetry, sealed [Shadow Score](https://github.com/DUBSOpenHub/shadow-score-spec/blob/main/SPEC.md) results, and [Fleet Scorecard](https://github.com/DUBSOpenHub/fleet-scorecard-spec/blob/main/SPEC.md) run decisions.



```text

You

  ↓

Agent Orchestra

  ↓

Terminal Stampede commander panes

  ↓

Bounded sub-agent groups

  ↓

Collaboration bus + live telemetry

  ↓

Shadow Score + Fleet Scorecard + final synthesis

```



Use it for final release reviews, architecture audits, migration plans, repo-readiness passes, and high-stakes implementation design where one model answer is not enough.



---



## ⚡ Quick Start



### Install from GitHub



```bash

curl -fsSL https://raw.githubusercontent.com/DUBSOpenHub/agent-orchestra/main/quickstart.sh | bash

```



The quick installer:



1. Checks for required local tools.

2. Verifies `python3` and `git` are available.

3. Installs the Stampede launcher, monitor, Fleet Scorecard, and Agent Orchestra helper launcher to `~/bin/`.

4. Runs the local smoke gate.

5. Leaves the repo ready for Agent Pulse observability and fleet validation.



### Install from a local clone



```bash

git clone https://github.com/DUBSOpenHub/agent-orchestra.git

cd agent-orchestra

./install.sh

```



### Run



```text

agent conductor on ~/dev/my-repo : evaluate auth architecture and propose fixes

agent conductor premium max on ~/dev/my-repo : final release readiness review

agent conductor standard small on ~/dev/my-repo : review docs and onboarding gaps

agent-orchestra-pulse

```



Agent Orchestra uses the Agent Conductor command surface for fleet missions and Agent Pulse for live visibility.



---



## 🧭 Command Reference



| Command | Purpose |

|---|---|

| `agent conductor` | Ask only for missing mission/repo; default to Premium Max + Agent Pulse + Stampede monitor |

| `agent conductor on REPO : MISSION` | Launch with Premium Max + Agent Pulse + Stampede monitor |

| `agent conductor premium max on REPO : MISSION` | Launch 5 premium commander groups with Agent Pulse + Stampede monitor |

| `agent conductor standard small on REPO : MISSION` | Launch 5 standard-tier commander groups with Agent Pulse + Stampede monitor |

| `agent conductor status [RUN_ID]` | Show concise stats, results, and collaboration counts |

| `agent conductor teardown RUN_ID` | Stop the underlying Stampede tmux session |

| `agent-orchestra-pulse` | Open Agent Pulse against this repo's fleet telemetry |

| `bin/fleet-scorecard --repo . --run-id RUN_ID` | Regenerate a Fleet Scorecard for a completed run |



> Note: scale words like `small`, `standard`, and `max` affect tier/policy, not commander count.

> Agent Orchestra follows the same five-commander fleet contract.



---



## 📊 Live Stats



The Stampede monitor and [Agent Pulse](https://github.com/DUBSOpenHub/copilot-cli-agent-pulse) read `.stampede/{run_id}/orchestrator-commentary.json` for concise, dashboard-safe stats:



```text

cmd 3/5 active · sub-agents 112 running / 480 done / 620 seen · q 0 · claimed 3 · results 2/5

collab p5 r18 i11 c8 b7

commander-004 launching_workers · squads 32/50 · sub-agents 160/250 · run 42 done 118 fail 0

```



In chat, Agent Orchestra uses swarm-style commentary: a short phase banner, commander progress table, totals, and a clear decision line (`wait`, `stop`, `synthesize partial`, or `relaunch`). Startup failures are called out as `failed_startup` instead of being described as merely still running.



Agent Orchestra distinguishes:



- **Running now** — live commander, squad lead, and worker telemetry.

- **Seen** — historical sub-agent launches recorded in the compatibility ledger.

- **Completed/failed** — terminal outcomes from commander telemetry.

- **Partial** — incomplete launches or degraded runs that must not be called success.



> 📡 **Best real-time view:** keep **Agent Pulse** open beside **Agent Orchestra** so commander status, sub-agent counts, recent launches, and telemetry confidence stay visible while the run is active.



---



## 👻 Shadow Score



Agent Orchestra creates a sealed [Shadow Score](https://github.com/DUBSOpenHub/shadow-score-spec/blob/main/SPEC.md) envelope before launch. Commanders and sub-agents see only the seal hash, never the hidden criteria. The orchestrator verifies the seal and evaluates final bundles only after the run completes.



Shadow Score focuses on:



| Dimension | Question |

|---|---|

| Requirement coverage | Did the final synthesis answer the mission? |

| Collaboration quality | Did commanders review, improve, and adopt useful findings? |

| Evidence quality | Are claims backed by files, logs, tests, or ledgers? |

| Validation impact | Did the run improve measurable confidence? |

| Synthesis usefulness | Did it combine the best work instead of picking vibes? |



---



## 📋 Fleet Scorecard



Every run also gets a [Fleet Scorecard](https://github.com/DUBSOpenHub/fleet-scorecard-spec/blob/main/SPEC.md)

overlay. The launcher follows the Fleet Scorecard Spec by sealing a lightweight

four-question rubric before commanders start, then teardown verifies the seal and

writes `.fleet-scorecards/{run_id}/scorecard.md`.



Fleet Scorecard answers:



| Question | Purpose |

|---|---|

| What changed? | Captures repo diffs, artifacts, decisions, and useful knowledge |

| What won? | Names the best commander output, idea, or recommendation |

| What failed? | Preserves partials, missing bundles, weak evidence, and teardown caveats |

| Would I run it again? | Gives a clear rerun decision and next-run modification |



This is intentionally different from cost accounting. It is the final

human-readable judgment layer that turns many commander outputs into one

repeatable run decision.



Agent Orchestra is the first **FSS-L4 reference implementation**: run card,

sealed rubric, evidence index, automatic teardown emission, and source run link.



Regenerate a scorecard manually:



```bash

bin/fleet-scorecard --repo . --run-id run-YYYYMMDD-HHMMSS

```



---



## 🏗️ Project Structure



```text

agent-orchestra/

├── agent-pulse-current/                  # Agent Pulse observability source

├── .fleet-scorecards/                    # Ignored Fleet Scorecard overlays

├── run-artifacts/                        # Local fleet run artifact corpus

├── skills/SKILL.md                       # Copilot CLI skill source

├── agents/                               # Worker, commander, and merger agent definitions

├── schemas/                              # Bundle and collaboration record schemas

├── bin/                                  # Terminal Stampede and Agent Orchestra launchers

├── tests/smoke.sh                        # Local smoke gate

├── install.sh                            # Local installer

├── quickstart.sh                         # One-command installer

├── scripts/activate-security.sh          # Repository security activation helper

├── AGENTS.md                             # Contributor/agent guide

└── archived-workflows/root/workflows/    # CI and CodeQL workflow templates

```



---



## 🧪 Validate Locally



```bash

bash tests/smoke.sh

```



The smoke test checks shell syntax, hardened runtime markers, fleet metadata,

commander bundle/schema integrity, Fleet Scorecard generation, Agent Pulse

import/poll behavior, and workflow activation safety.



---



## 🤝 Contributing



See [CONTRIBUTING.md](CONTRIBUTING.md). The short version:



1. Keep the orchestration flow self-contained.

2. Preserve sealed Shadow Score isolation.

3. Preserve Fleet Scorecard seal verification and the four-question output.

4. Use **sub-agents** in user-facing language.

5. Keep live commentary concise.

6. Run `bash tests/smoke.sh` before opening a PR.



## License



MIT — see [LICENSE](LICENSE).



---



🐙 Created with 💜 by [@DUBSOpenHub](https://github.com/DUBSOpenHub) with the [GitHub Copilot CLI](https://docs.github.com/copilot/concepts/agents/about-copilot-cli).



Let's orchestrate! 🎼✨