Skip to main content

What Happens During Execution


1. Entry

After Payment, you are redirected to
/execution?sessionId=...&platform=...&signature=...
(or without signature when paying with Dodo).
  • The Execution page loads
  • After a short “Preparing…” delay (~2 seconds)
  • Calls POST /api/execute with:
    • sessionId
    • transactionSignature (if Solana)
    • platform

2. Execute API

The /api/execute handler:
  1. Loads workflow context
    • From database (or in-memory) using sessionId
    • If missing, may rebuild from Workflow table or fail with a clear error
  2. Validates
    • Ensures selectedPlatform exists
    • Ensures refinedRequirements exist
    • If Solana payment, may verify transaction on-chain
  3. Resolves agents
    • From platformOptions.agents_required
    • Or defaults (ProductManager, FrontendDev, BackendDev)
    • For Android / iOS, DevOpsEngineer usually included
    • ReviewAgent always runs after main agents
  4. Runs the X402 workflow
    • Calls executeX402Workflow
    • Inputs: platform, requirements, sessionId, agent list
  5. Saves the result
    • Writes executionResult to workflow:
      • agents_executed
      • deliverables
      • review_passed
      • review_issues
    • Returns the result

3. X402 Workflow (server-side)

Inside executeX402Workflow:
  1. Platform config
    • Selects template (Website, Web Application, Android, iOS)
    • Gets expected folder structure and file list
  2. Platform deliverables
    • generatePlatformDeliverables creates real file content:
      • PRD
      • PLAN
      • Specs
      • Prompts
      • Infra
    • Files assigned to agents and generated via AI
  3. A2A setup
    • Initializes A2A (Agent-to-Agent) for the session
    • Sends “Execution started”
    • Shares requirements with all agents
  4. Agent run
    • Agents run in parallel:
      • ProductManager
      • FrontendDev
      • BackendDev
      • DevOpsEngineer
      • ToolManager (as applicable)
    • Each agent:
      • Receives requirements
      • Produces outputs
      • Shares execution_result via A2A
    • ReviewAgent does not run here
  5. Aggregation
    • Platform deliverables are merged
    • Deduplication may regenerate identical files
  6. Folder structure
    • Adds project_structure deliverable
    • Example: FOLDER_STRUCTURE.md
  7. Review
    • ReviewAgent checks:
      • Required files exist
      • Each file has path and non-empty content
    • Returns:
      • review_passed (true/false)
      • review_issues (list)
  8. Response
    • Returns:
      • agents_executed
      • deliverables
      • integration_points
      • review_passed
      • review_issues
      • execution_time_seconds

4. Projects Generate (ZIP)

When Execute API returns success and review_passed = true:
  • Execution page calls POST /api/projects/generate
  • Inputs:
    • sessionId
    • executionResult
That API:
  • Builds project file tree
  • Writes files to disk or in-memory
  • Creates a ZIP
  • Stores ZIP in ZipFile (and/or Workflow.zipFile)

5. Execution Page After Completion

  • Review
    • “Review Passed” or “Review Failed”
    • Shows review_issues if failed
  • Structure
    • Renders Project Structure preview
  • ZIP
    • Progress bar during:
      • “Generating ZIP file…”
      • “Preparing your document…”
    • Shows “Download Project ZIP” when ready
  • GitHub
    • If OAuth configured:
      • Connect to GitHub
      • Upload to GitHub
      • Create repo and push files

Agent Coordination


Roles in Execution

AgentRole
X402AgentOrchestrator; shares requirements via A2A; coordinates run
ProductManagerPRD, user stories, acceptance criteria, feature specs
FrontendDevFrontend components, pages, styles, specs
BackendDevAPI, data models, DB schema, auth
DevOpsEngineerDocker, CI/CD, deployment, infra
ToolManagerIntegrations, .env.example, configs, folder structure
ReviewAgentPost-run review; sets review_passed and review_issues
CEO appears in UX copy, but real orchestration is done by X402Agent.

A2A (Agent-to-Agent)

  • Initialize
    • A2A initialized at start of X402 run
  • Notifications
    • “Execution started”
    • “Starting <Agent>”
    • “<Agent> completed”
  • Data sharing
    • Requirements shared to all agents
    • Agents share execution_result when done
  • Parallel run
    • Agents run via Promise.all
    • No fixed sequence

Integration Points

Inferred from agents run:
  • Frontend–Backend
  • Backend–AI
  • ProductManager–Frontend
  • DevOps–Backend
Reported in execution result.

Who Produces What

  • Platform-specific content
    • Produced by generatePlatformDeliverables
  • Parallel agent run
    • Provides status and high-level outputs
  • Final deliverables
    • Platform deliverables
    • FOLDER_STRUCTURE.md
    • REVIEW_REPORT.md

Progress Tracking


On the Execution Page

PhaseWhat you see
Preparing“Preparing… Setting up your project”
Executing“Executing Workflow” + messages
Execution complete“Execution complete!”
Review“Review Passed” or “Review Failed”
Generating ZIPProgress bar (0–100%)
Ready“ZIP file ready!” + download button

Progress Messages (in order)

  1. “Validating payment…”
  2. “CEO Agent: Analyzing request and routing to X402Agent…”
  3. “X402Agent: Orchestrating simultaneous multi-agent execution…”
  4. “Executing agents simultaneously: ProductManager, FrontendDev, BackendDev…”
  5. “ReviewAgent: Reviewing deliverables for quality and uniqueness…”
Note
  • /api/execute is a single HTTP request
  • Messages are client-side placeholders
  • Server work runs in one pass
  • ZIP progress reflects /api/projects/generate

If Review Fails

  • “Review Failed” shown
  • review_issues displayed
  • ZIP generation not started
  • Download button hidden

Expected Timelines


By Platform (typical)

PlatformTypical execution time
Website~45 s
Web Application~75 s
Android~100 s
iOS~100 s

Additional Time

StepTypical duration
Preparing (client)~2 s
Execute API~45–100+ s
Projects Generate (ZIP)~5–30+ s
Client UX placeholdersA few seconds

End-to-End (Payment → Download)

  • Website: ~1–2.5 min
  • Web Application: ~1.5–3 min
  • Android / iOS: ~2–3.5 min

What Affects Duration

  • Platform
  • Deduplication
  • ReviewAgent overhead
  • AI model and RPC latency
  • ZIP size and file count