Source:
docs/manual/scripts-demos.mdThis page is generated by
site/scripts/sync-manual-docs.mjs.
Scripts and Demos
Demo scripts validate real agent workflows using simulated tools and approvals. They exercise the full execution path — Temporal workflow, LLM decision loop, tool routing, approval gates, and cost tracking — without requiring external service dependencies. Use them to verify that the runtime works end-to-end after configuration changes or upgrades.
This document is for developers running validation checks and operators verifying deployment health.
Prerequisites: Running Temporal dev server, seeded tool registry (version v2.0.7 or later), Postgres available.
Available Demos
ArgoCD ApplicationSet Alert Demo
Script: scripts/argocd_applicationset_alert_demo.sh
What it demonstrates: An agent responding to an ArgoCD application health alert. The agent inspects the application status, identifies the failing component, and recommends a remediation action. Exercises the http_get and bash_exec simulated tools with approval gates.
How to run:
bash scripts/argocd_applicationset_alert_demo.sh
What to look for: The workflow should complete with a halt decision after 3-4 steps. Check that approval signals were sent and acknowledged, and that the cost summary shows non-zero token usage.
Introduced in: Phase 3 (tool execution and approvals)
Ticket New Runtime Demo
Script: scripts/ticket_new_runtime_demo.sh
What it demonstrates: An agent processing a new support ticket. The agent triages the ticket, gathers context from simulated APIs, and drafts a response. Exercises multi-step tool chaining with the sim_ticket_read, sim_search, and sim_draft tools.
How to run:
bash scripts/ticket_new_runtime_demo.sh
What to look for: The workflow should show a clear triage → research → draft sequence in the decision log. Verify that each tool call has valid schema-conformant arguments.
Introduced in: Phase 4 (memory and context)
Incident Cost Regression Demo
Script: scripts/incident_cost_regression_demo.sh
What it demonstrates: An agent investigating a cost regression incident. Exercises cost tracking, model preference lookups, and the cost query CLI. Uses simulated tools that return cost metrics.
How to run:
bash scripts/incident_cost_regression_demo.sh
What to look for: The cost summary at the end should reflect the models used. Check that the agent selected the preferred model from the model preferences configuration.
Introduced in: Phase 6 (models and cost tracking)
Phase 7A Smoke Test
Script: scripts/phase7a_smoke_test.sh
What it demonstrates: End-to-end smoke test covering supervisor patterns, graph workflows, and approval flows. Validates that all Phase 7 features (supervisor broadcast, delegate, debate) work together.
How to run:
bash scripts/phase7a_smoke_test.sh
What to look for: Each supervisor pattern should complete without errors. The script prints pass/fail for each test case.
Introduced in: Phase 7A (supervisor patterns)
Writing a New Demo Script
To create a new demo:
- Create a script in
scripts/following the naming convention<topic>_demo.sh. - Use simulated tools (prefixed
sim_) to avoid external dependencies. Simulated tools force approval gates so the demo exercises the approval flow. - Seed the registry with the required version before running the workflow.
- Include clear output messages indicating what the demo is testing and whether each step passed.
- Exit with a non-zero code on failure so CI can detect regressions.
Example skeleton:
#!/usr/bin/env bash
set -euo pipefail
echo "=== My Demo: testing feature X ==="
# Seed registry
go run ./cmd/seed-registry --version v2.0.7
# Run workflow
go run ./cmd/run --prompt "Test feature X" --max-steps 4
# Query result
go run ./cmd/query --workflow-id <id> --query current_state
echo "=== Demo complete ==="
Notes
- Simulated tools are prefixed
sim_and force approval gates. - Registry must be seeded with version
v2.0.7or later for all demos. - Demo scripts are not run in CI by default. Add them to the CI workflow if you want automated validation.