Operate the simulator with Claude Code

The optional ireasoning-simulator skill lets you operate SNMP Agent Simulator using natural language in Claude Code. You can search the device library, create agents, control labs, save and run scenarios, and explain or export run results. It uses the simulator REST API, so the same labs and results are visible in the desktop application.

Requirements

Install and sign in to a current Claude Code release that supports skills. An active Claude account with available usage is required. Install Python 3.9 or later on the computer running Claude Code. Use python3 on macOS or Linux, or py -3 on Windows. The skill uses only the Python standard library; no additional Python packages are required.

Use a simulator distribution containing skills/ireasoning-simulator or ireasoning-simulator-skill.zip. The simulator and its REST service must be running with the usual simulator license. This integration targets Claude Code terminal sessions; it does not provide an embedded chat window or a Claude Desktop MCP connection.

Install the skill

For personal use, place the ireasoning-simulator folder inside your user account's .claude/skills directory. On Windows this is %USERPROFILE%\.claude\skills; on macOS or Linux it is ~/.claude/skills. If using the ZIP, extract it into that skills directory. The final layout must contain .claude/skills/ireasoning-simulator/SKILL.md, with scripts and references alongside it, rather than a second nested ireasoning-simulator folder.

Alternatively, copy the supplied folder. Open a terminal in the simulator distribution directory, which contains the skills folder, and run the commands for your platform.

macOS or Linux:

mkdir -p ~/.claude/skills
cp -R skills/ireasoning-simulator ~/.claude/skills/

Windows PowerShell:

New-Item -ItemType Directory -Force "$HOME\.claude\skills"
Copy-Item -Recurse .\skills\ireasoning-simulator `
  "$HOME\.claude\skills\"

For a skill used only in one project, install the same folder under that project's .claude/skills directory instead. Inspect or back up an existing copy before replacing it. Start a new Claude Code session after installing the skill, or use the skill reload command supported by your Claude Code release.

Connect to the simulator

In SNMP Agent Simulator, choose Tools > Start REST API Service. Check the bind address and port, then click Start. For Claude Code running on the same computer, the skill connects to http://127.0.0.1:18080 by default. Starting the REST service does not start the lab agents.

For a simulator on another computer, start the REST service on a reachable interface and tell Claude its explicit service URL, for example http://192.168.1.20:18080. Use your existing network and HTTPS configuration as appropriate. The skill does not change server settings or switch to another endpoint if the specified service is unavailable.

Make a request in Claude Code

Invoke the skill explicitly with /ireasoning-simulator followed by your request. Claude Code can also select it when you ask naturally about iReasoning Simulator. Begin by listing the available labs so that Claude can resolve the real lab and agent IDs.

/ireasoning-simulator Show my labs and their agent states.

Example requests:

• In Sample Simulators, show the agents and start only cisco.

• Create a new lab with ten Cisco Catalyst 9300 agents from the device library on 127.0.0.1, using ports 1161 through 1170. Leave them stopped.

• In this lab, create a 30-second scenario that starts the selected agent, applies a five-second outage at 10 seconds, and checks for requests after recovery. Save the scenario before running it.

• Send one linkDown trap to 127.0.0.1 port 1162 at three seconds into the scenario. Leave the agent's normal trap recording behavior unchanged.

• Explain why the selected run failed, including failed checks and observed counts, and export its JUnit report to my chosen directory.

Specify the lab, agent names, timing and destination when relevant. If names match more than one lab or agent, choose the exact ID when Claude asks. Device-library requests use the currently installed catalog; an entry with a missing recording cannot be instantiated. Several agents on one IP address with different UDP ports are separate endpoints, not separate discoverable IP addresses.

The skill uses the existing simulator operations and their state requirements. Structural changes require a stopped lab. Starting a scenario resets its baseline and stops its agents when the run finishes. Check these effects when asking Claude to run a scenario on an existing lab. Claude Code's normal tool permissions still apply.

Understand progress and results

REST operations run asynchronously. The helper normally waits for completion and reports operation and run IDs. For a long operation, ask Claude to submit it without waiting and check its operation ID later. A timeout or interrupted command does not cancel the server operation. Ask Claude to inspect or resume that operation rather than submitting the same action again with a new idempotency key, especially when sending notifications.

A SUCCEEDED run without checks means its actions completed; it does not prove that a network manager received traffic or recovered correctly. Checks need real observations from your SNMP manager, poller or receiver. Sending a UDP trap only establishes local transmission; an INFORM acknowledgement provides receiver evidence. Reports identify incomplete evidence and dropped display events. Ask Claude to export the full report when the event history is large.

Agent trap recordings, threshold rules and scripts continue to send their normal notifications independently of scenario Send trap / INFORM actions. The skill can explain available reports and events; it does not provide CPU or thread profiling or undo already transmitted notifications.

Troubleshooting and removal

If Claude cannot connect, check that the REST service is started and that the selected host and port are correct. If the skill is not found, verify the SKILL.md location and start a new Claude Code session. If Python is missing, install Python 3.9 or later and check python3 --version or py -3 --version. Claude account or credit errors must be resolved in Claude; restarting the simulator does not resolve them.

An agent that cannot start may have an endpoint conflict, a license capacity limit or a recording that requires trusted scripts. A headless service disallows recording scripts unless explicitly started with --trusted-scripts. The skill does not bypass this policy. Keep actual communities and SNMPv3 passwords in private input files; do not paste them into chat or request that they be printed.

To update the skill, back up any local customizations and replace its installed folder with the new distribution copy. To uninstall it, remove only the installed ireasoning-simulator skill folder. This leaves the simulator, labs, scenarios and reports unchanged.