Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Autonomous Workflow Architect

    1

    Design reliable end-to-end agentic workflows with explicit states, approvals, verification, recovery, and bounded autonomy.

    $9.99

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Autonomous Workflow Architect

    Autonomous Workflow Architect

    Example session with this skill installed

    Design a workflow that receives a customer refund request, checks order eligibility, drafts a response, and may issue a refund. Include states, tools, human approval gates, idempotency, audit events, retries, failure recovery, and tests. Assume payment-system access is not available, and do not execute or imply that any refund occurs.

    • Read your context and instructions
    • Compiled the autonomous workflow architect

    Refund request workflow design

    Goal and boundary

    Process a refund request into a verified decision and customer response. The design below does not issue refunds; payment-system access is not available in this scenario. Any future refund action requires an authorized integration and explicit approval policy.

    State machine

    StateAction / evidenceTransition
    ReceivedValidate request ID, customer reference, order ID, and consented contextMissing or invalid data → Needs information; otherwise → Eligibility check
    Eligibility checkRead order date, item, delivery/return status, prior refunds, policy version, and currency through authorized read-only toolsUnclear/conflicting → Human review; eligible → Draft decision; ineligible → Draft explanation
    Draft decisionGenerate reasoned outcome with citations to order facts and applicable policy→ Approval required
    Approval requiredPresent amount, reason, evidence, policy, and irreversible side effect to authorized reviewerApprove → Ready for action; reject/edit → Revised decision
    Ready for actionRecheck order and duplicate-refund state; require idempotency key and amount capAuthorized payment tool present → Execute; otherwise → Manual handoff
    ExecutingCall payment API once with idempotency key; record response IDSuccess → Verify; timeout/unknown → Reconcile, do not retry blindly
    VerifyRead authoritative payment/refund statusConfirmed → Completed; mismatch → Escalate
    CompletedSend approved customer message and audit final statusTerminal

    Tool contracts and controls

    Order lookup (read only): order ID, line items, status, currency, timestamps, prior refunds; reject ambiguous matches.

    • Policy lookup (read only): versioned policy and effective date; surface conflicts for human review.

    Refund API (side effect): unavailable in this design. If later enabled, require least privilege, strict amount bounds, idempotency key, approval record, and authoritative status verification.

    • Messaging (side effect): send only after outcome is verified and an approved template is selected.

    Recovery and failure handling

    Timeout after refund call: query authoritative payment status using the same request and idempotency key; never issue a second refund until reconciliation. Duplicate request: deduplicate on request ID and order/amount key. Policy conflict, mismatch, unavailable order data, or suspected fraud: stop automation and route to a designated reviewer. Log input references, policy version, tool calls, approvals, outcomes, timestamps, and error codes without storing unnecessary payment details.

    Acceptance tests

    1. Missing order ID pauses for information and creates no side effect.
    2. Conflicting eligibility evidence routes to review.
    3. Repeated delivery of the same event does not create a second refund.
    4. Reviewer rejection blocks execution.
    5. API timeout triggers reconciliation rather than blind retry.
    6. Customer notification is withheld until authoritative success is confirmed.
    7. No payment tool available results in a manual handoff and an explicit “not executed” status.

    Deployment gate: Security, privacy, finance, and customer-support owners must approve tool permissions, policy mapping, amount limits, audit retention, and escalation coverage before implementation.

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

    About this skill

    What this skill does

    Autonomous Workflow Architect designs, audits, debugs, and improves AI-driven processes that move from an objective to verified completion. It converts requirements into workflow graphs with explicit states, tool contracts, evidence requirements, approval gates, side-effect controls, checkpoints, recovery paths, escalation, termination, observability, cost, security, and governance.

    Workflow

    Define success and triggers; identify inputs, systems, actors, and authoritative state; separate deterministic steps from AI judgment; map states and transitions; define tool contracts and permission boundaries; place approvals before consequential side effects; specify verification and idempotency; design retries, timeouts, recovery, compensation, and human escalation; establish observability and cost limits; test failure scenarios and document operations.

    Deliverables

    Workflow graph, state machine, tool/API contracts, approval map, data-flow and permission plan, verification checklist, retry and recovery design, failure-mode analysis, observability plan, test cases, and runbook.

    Limits

    It never invents tool access, approvals, external state, credentials, deployment, workflow completion, or successful side effects. It designs workflows but executes or deploys only when the host supports and explicitly authorizes those actions.

    Workflow designs depend on complete requirements, reliable tools, authoritative state, clear policies, and implementation quality. The architecture cannot guarantee correct model decisions, service availability, security, compliance, or outcomes. Consequential actions need authorized integrations, bounded permissions, human approval where appropriate, and tested verification and recovery.

    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 any payment.

    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

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 1 month ago

    Permissions

    Read Files
    Write Files

    File Scopes

    *.md
    *.yaml
    *.json

    Use only requirements, schemas, and documentation provided or explicitly authorized. Read supplied workflow materials and write diagrams or specifications to the designated workspace. This skill designs workflows but does not grant tool access, execute side effects, deploy automation, or bypass approval gates. Require least privilege, human approval for consequential actions, idempotency, and verification before implementation.

    Listed1 month ago

    What's inside

    Frequently Asked Questions

    It produces workflow architecture, diagrams, contracts, test plans, and runbooks. It does not deploy or execute automation unless the host supports those tools and the user authorizes the specific implementation.

    Use idempotency keys, authoritative status reconciliation, bounded retries, and explicit recovery or compensation paths. Never retry an uncertain side effect blindly; escalate when state cannot be verified.

    Place approval before irreversible or consequential actions such as payments, account changes, external messages, data deletion, or access changes. Show the reviewer the proposed action, evidence, scope, and likely impact, then verify the result afterward.

    New in Agents & Orchestration