Relay Orchestra

    1

    - Review a large codebase by subsystem or specialist perspective.

    Free

    1 useSecurity scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Relay Orchestra

    Example session with this skill installed

    Use Relay Orchestra to take a booking product from a short brief to a backend ready for a private beta. Customers should be able to find available times, book a service, and cancel; businesses should manage services and availability.

    Coordinate product scope, API and data design, implementation, independent review, and deployment preparation. Keep payments, mobile apps, and advanced analytics for a later release. Protect against double bookings and access to another business's data.

    Keep this as a full live session so I can change priorities while work continues. Use the existing project and available tools. Prepare the staging release, smoke checks, and rollback instructions; stop before creating paid resources or deploying externally so I can review the release package.

    • Read your context and instructions
    • Compiled the relay orchestra

    Booking product: private beta release package

    Illustrative scenario for a fictional product. The deliverables below show an example handoff, not an executed deployment or verified test run.

    Product scope

    The first release covers the complete booking journey: discover a service, choose an available time, reserve it, and cancel if needed. Business owners manage their own services and availability. Payments, native apps, and advanced analytics remain outside this release.

    One coordinated delivery

    AreaExample deliverable
    ProductA short beta brief, user journeys, and acceptance criteria
    BackendAuthentication, business accounts, service catalog, availability, and booking API
    DataSchema and migrations, tenant ownership rules, and database-level protection against overlapping bookings
    ReviewIndependent checks of authorization, booking races, cancellation, and retry behavior
    OperationsContainer setup, environment template, health checks, staging instructions, and rollback runbook

    API contract

    ActionEndpointExpected behavior
    Browse servicesGET /servicesReturn the selected business's public services
    Find a timeGET /availabilityReturn bookable times in the requested range
    BookPOST /bookingsCreate one reservation; reject an unavailable time with 409
    View bookingsGET /bookingsShow only reservations the current account may access
    CancelPOST /bookings/{id}/cancelApply the cancellation policy; a retry must not create another change

    Coordination and verification

    Product and architecture research can run in parallel. One implementation owner integrates their findings; independent reviewers inspect the result before release preparation. Additional writers require approved isolation or work sequentially.

    The release evidence must cover two customers trying to reserve the same time, cross-business access attempts, repeated booking requests, cancellation, and a clean database migration. The final handoff should link each check to its actual result and disclose failures or untested behavior.

    Release handoff

    1. Review the proposed environment, configuration, migration order, and operating cost.
    2. After deployment is authorized, apply the release to staging and check health, authentication, and the full booking journey.
    3. Record the deployed revision and actual staging URL. If checks fail, follow the rollback runbook and report the remaining blocker.

    The coordinator returns one release report with the delivered scope, verification evidence, deployment status, and next product decisions. This illustrative example creates no cloud resources.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Coordinate native AI agents for coding, research, audits, and migrations.

    Relay Orchestra is an open-source Agent Skill. It gives workers focused tasks, manages dependencies and file ownership, and checks their results before returning one synthesis. Released under the MIT license.

    What it helps with

    • Research a technical integration from several angles, then turn findings into a client/backend handoff.
    • Review a large codebase by subsystem or specialist perspective.
    • Combine documentation research, implementation, and independent verification.
    • Coordinate changes across modules or repositories while respecting existing work and approval boundaries.

    How it works

    1. Define the objective, constraints, worker roles, and expected evidence.
    2. Use a lightweight one-shot run for bounded work with one writer and read-only workers.
    3. Use Full coordination when multiple writers, worktrees, changing requirements, or continuity across turns need explicit control.
    4. Review worker artifacts and side effects, perform appropriate checks, and report the result with remaining limitations.

    For concurrent writers, Relay plans isolated Git worktrees and seeks approval before creating them. Without isolation, writers run sequentially. File ownership and integration review reduce conflicts; they do not guarantee that every merge or agent result is correct.

    Example workflow: launch a product backend

    Take a booking product from a short brief to a private beta release package. Coordinate product scope, API and database design, backend implementation, independent review, and deployment preparation. Keep one owner responsible for integrating the work and reporting verified results. The demo below is an illustrative fictional scenario; it uses no private chat content and does not claim an actual deployment.

    Invocation and control

    Invoke the installed skill with a concrete task. In Codex: $relay-orchestra Review the current changes and verify important findings. Request a full live session when you want to keep steering related work across turns. Natural-language instructions can set effort, scope, and worker limits.

    Optional Auto applies the router to suitable later tasks in the same chat. It stays idle between tasks and does not transfer to unrelated chats. auto off disables that future preference; stopping an active live session is a separate instruction.

    Requirements and limits

    The skill itself has no runtime dependencies. Parallel execution, background work, notifications, interruption, and automatic resumption depend on the host client and version. Where native agents are unavailable, Relay offers sequential work or dispatch-ready briefs. Delegated workers consume additional model usage. Relay grants no new permissions and makes no guarantee of bug-free output.

    Source, installation instructions, and support: https://github.com/SDA-31/relay-orchestra

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Agents & Orchestration.

    Host capabilities determine native subagents, concurrency, background execution, notifications, automatic resumption, and interruption. A SKILL.md-compatible client alone does not guarantee full live behavior. Auto is chat-scoped and depends on retained context. Lite permits one writer; concurrent writers need approved isolation or run sequentially. Additional workers consume model usage. Verification reduces risk but cannot guarantee correctness.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before buying a skill.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    1 use

    Times used to date

    Free to use

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 4 days ago

    • Free with an Agensi account

    Nothing to install. It runs in your AI agent.

    No runtime dependencies. Parallel execution needs native subagents; live/background support varies by client. Installation options and OS tools are documented in INSTALL.md.

    Permissions

    Terminal / Shell
    Read Files
    Write Files

    Tool access depends on the requested task and the host's existing permissions. Coding tasks may use shell, file reads/writes, Git, and approved worktrees; research may use available browsing tools. No third-party service account or API key is intrinsically required. Relay does not expand permissions.

    Listed5 days ago

    What's inside

    Frequently Asked Questions

    An Agent Skills-compatible client is needed for normal skill discovery. Native subagents enable parallel execution; live background behavior depends on the client and version. Relay itself has no runtime dependencies. Node.js/npm is used by the optional skills CLI installer; standalone installers use Python 3.7+ with the documented OS tools.

    Relay assigns file ownership and reviews changes before integration. Concurrent writers use approved isolated Git worktrees; without isolation, writers run sequentially. Approved isolated writers may overlap files under an explicit integration plan. These are coordination rules, not a guarantee that merge conflicts or bugs cannot occur.

    Yes. Use natural language for depth, speed, scope, and resource limits. An exact worker count is both a target and a cumulative ceiling, normally excluding the coordinator. Up to N is a ceiling; at least N is a floor. Relay works within the host's capacity and may schedule waves.

    The coordinator inspects worker artifacts, evidence, ownership, and side effects, then performs appropriate checks before synthesis. Independent reviewers are used when they add value. The final report distinguishes verified results, remaining uncertainty, and blockers; verification cannot guarantee correctness.

    Yes. Relay Orchestra is released under the MIT license. Source, installation instructions, references, and examples are available at https://github.com/SDA-31/relay-orchestra. Delegated workers still consume usage under your AI client's plan.

    A one-shot run finishes a bounded task and settles its workers in the same response. Full live coordination stays active for related follow-ups until its lifecycle is closed. Auto is an optional preference for suitable later tasks in the same chat; it launches nothing while idle and does not transfer to other chats. Say auto off to disable that preference, and stop the live session separately if one is active.

    New in Agents & Orchestration

    Free