Local Agent Prompt Test¶

This quickstart runs the demo the easiest way: install OpenCode, open this repository in OpenCode, select the network-troubleshooter agent, paste one test prompt, and watch the agent execute a Repair Action Workflow with simulated network-device responses.

You do not need Splunk, Docker, Webex, RADKit, or Python for this path. The prompt points the agent at a checked-in RAW test bundle, so the agent uses canned CLI output instead of touching a real device.

For the other setup paths:

Goal	Start here
Connect the demo to your own routers, RADKit MCP, Splunk, and Webex	Lab Environment
Create your own Fault Signature, Repair Action Workflow, and Remediation Guide artifacts	Artifact Authoring
Maintain the KB vault	Knowledge Base Curator

1. Open a Terminal in This Repo¶

From a command prompt or terminal:

cd c:/src/fault-intelligence-as-code

If you cloned the repo somewhere else, use that path instead.

2. Start OpenCode¶

For the browser UI:

opencode web

OpenCode will launch or print a local browser URL, usually http://localhost:4096.

If you prefer the terminal UI, run:

opencode

3. Select the Network Troubleshooter Agent¶

In OpenCode, start a new chat and select the network-troubleshooter agent.

This matters because the test-mode behavior is defined on that agent. In test mode, it must read the test bundle, simulate CLI responses, avoid RADKit, apply scripted approvals, and write test-run artifacts under logs/test-runs/.

4. Paste One Test Prompt¶

Pick either scenario and paste the full prompt into the network-troubleshooter chat.

Option A: AD000002 Admin Shutdown¶

This is the primary demo scenario. It simulates a BGP neighbor that is down because shutdown is present in the IOS XR BGP neighbor configuration. The workflow confirms the state, applies the scripted no shutdown repair, and verifies the neighbor returns to Established.

Run the RAW test bundle for AD000002 / RAW000002 in test mode.

Simulate network-device responses from the test bundle. Do not use RADKit, Splunk, Webex, Docker, or the alert simulator script. Use the network-troubleshooter test-mode behavior and write the generated test artifacts under logs/test-runs/.

{
  "alert_def_id": "AD000002",
  "device_hostname": "xr-43",
  "mode": "strict",
  "test_bundle_path": "intelligence-artifacts/AD000002-bgp-neighbor-admin-shutdown-xr/tests/RAW000002-BGP_NEIGHBOR_ADMIN_SHUTDOWN_REPAIR.tests.yml",
  "test_name": "resolve_via_no_shutdown",
  "webex_notify": false
}

Option B: AD000003 Maximum Prefix¶

This scenario simulates a BGP neighbor reset caused by maximum-prefix enforcement. The workflow uses canned device responses to identify the recovery path, run the scripted rollback/clear actions, and verify recovery.

Run the RAW test bundle for AD000003 / RAW000003 in test mode.

Simulate network-device responses from the test bundle. Do not use RADKit, Splunk, Webex, Docker, or the alert simulator script. Use the network-troubleshooter test-mode behavior and write the generated test artifacts under logs/test-runs/.

{
  "alert_def_id": "AD000003",
  "device_hostname": "xr-edge-01",
  "mode": "strict",
  "test_bundle_path": "intelligence-artifacts/AD000003-bgp-max-prefix-adjchange-xr/tests/RAW000003-BGP_NEIGHBOR_MAX_PREFIX_LIMIT_EXCEEDED_REPAIR.tests.yml",
  "test_name": "resolve_default_vrf_recovery",
  "webex_notify": false
}

5. Watch the Remediation Flow¶

The agent should move through the same major stages as a live alert, but with simulated device responses:

Stage	What to watch for
Test-mode detection	The agent notices `test_bundle_path` and avoids live RADKit calls.
Artifact loading	The agent loads the matching FS, RAW, RG, and test bundle.
Validation	The RAW validation steps consume canned CLI output from the test bundle.
Approval handling	Config actions use scripted approvals from the test bundle.
Remediation	The agent follows the RAW action path and reports the terminal outcome.
Summary	The agent emits a session summary with the step path, variables, and result.

For this quickstart, a successful run should not ask you to start Splunk, configure Webex, run Docker, or connect to a router.

6. Inspect the Generated Artifacts¶

After the run finishes, look under:

logs/test-runs/

Each agent test run creates a timestamped folder. The exact contents depend on the scenario and agent path, but expect artifacts like:

Artifact	Purpose
`session.md`	Human-readable troubleshooting transcript and decision log.
`result.json`	Structured test result: outcome, step path, variables, expectations, and diffs.
`bundle/cli/`	Captured or simulated command evidence used by the workflow.
`bundle/report.html`	Troubleshooting bundle report, when generated.

If you use VS Code, open logs/test-runs/ in the Explorer and sort by modified time. From PowerShell, you can list the newest folders with:

Get-ChildItem logs\test-runs | Sort-Object LastWriteTime -Descending | Select-Object -First 5

7. Try More Test Paths¶

The prompts above run one happy-path test each. To run every canned path in a bundle, paste the same prompt but remove the test_name line.

For AD000002, the bundle includes paths for successful repair, non-admin shutdown escalation, transient recovery, failed recovery, external recovery, and approval denial.

For AD000003, the bundle includes paths for already-recovered sessions, rollback recovery, policy discovery failures, approval denial, and post-rollback recurrence.

8. Next Steps¶

Need	Doc
Connect to a real lab	Lab Environment
Create a new fault scenario	Artifact Authoring
Understand the agent architecture	Agents
Troubleshoot setup issues	Troubleshooting