Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add xorrkaz/cml-mcp --skill cmlgit clone --depth 1 https://github.com/xorrkaz/cml-mcpWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/xorrkaz/cml-mcp/cml)<a href="https://agentmods.dev/skills/xorrkaz/cml-mcp/cml"><img src="https://agentmods.dev/badge/skills/xorrkaz/cml-mcp/cml/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/xorrkaz/cml-mcp/cml"><img src="https://agentmods.dev/badge/skills/xorrkaz/cml-mcp/cml.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00141 | $0.03218 |
| Opus 5.5 | $0.00056 | $0.01287 |
| Sonnet 5.5 | $0.00028 | $0.00644 |
| Haiku 4.5 | $0.00014 | $0.00322 |
Grade A, and why
cml scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 5d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 114 lines — stays where its author put it; the contents beside it link to each section on GitHub.
CML
Build and operate CML labs through the CML MCP tools. The guidance below covers the conventions and pitfalls an agent will not infer correctly on its own. General networking knowledge (e.g., what OSPF is, how a switch works) is assumed and not repeated here.
IDs are opaque — copy them, never reconstruct them. When a tool returns an identifier (lab, node, interface, or link UUID), pass that exact value to later calls. Never retype an ID from memory or rebuild one from context; a single altered character makes the call fail, and the corrupted value can still look well-formed, so you cannot catch the error by inspecting it. If you are not certain you have the exact value, re-fetch it with the relevant listing tool (get_nodes_for_cml_lab, get_interfaces_for_node, get_all_links_for_lab) rather than guessing. One exception: send_cli_command/send_cli_config_command identify their node by label, not UUID — see Configuration.
Default approach: build incrementally, not in one shot
Build a lab by composing small, verifiable steps. Do not try to hand-assemble a complete topology object and import it in one call unless the user has explicitly provided a finished model.
Default sequence:
create_empty_lab(title, optional description/notes) → capture the returnedlab_id. Lab notes are rendered as Markdown in the CML UI, so record the topology's purpose / high-level description there — it's what a human sees when they open the lab.get_cml_node_definitionsonce, to learn validnode_definitionvalues for this server. Do not guess names likecsr1000v/iosv— servers differ in what's installed.- Call the tool and check the result. If it returns an error or an empty list, stop immediately and tell the user: the server returned no node definitions and you cannot proceed without them. Do not guess any
node_definitionvalue. - If the device type the user asked for isn't in the returned list, stop and tell them — name the closest available definitions and let them choose. Never silently substitute a different device type.
- If two or more definitions share the same vendor and device family as the requested device (matching the family name in the definition name or the description's product line — e.g. "a Cisco router" matching
iol-xe,iosv,csr1000v,cat8000v), list each one by its exact name and ask the user to choose before proceeding. Do not silently choose the first match. - Otherwise, use the single matching definition.
- For which definition to recommend as the default among installed options (e.g.
iol-xefor routers,ioll2-xefor switches), seereferences/node_selection_guidance.md.
- Call the tool and check the result. If it returns an error or an empty list, stop immediately and tell the user: the server returned no node definitions and you cannot proceed without them. Do not guess any
add_node_to_cml_labper device. Setlabel,node_definition, andx/y(see Layout). Each add auto-creates that definition's default interfaces.- Connect nodes (see Connecting nodes — this is the step most often done wrong).
- Apply configuration (see Configuration).
- Start nodes when the user wants the lab running. Before starting a node, read its current state with
get_nodes_for_cml_lab. If it is already BOOTED, skip the start call and report that. If the state cannot be read, report that and do not start it. To start an entire lab, usestart_cml_lab(lab_id). To start nodes in a specific order (e.g., core infrastructure before edge devices), callstart_cml_node(lab_id, node_id)sequentially, usingwait_for_convergence=trueon each node before starting the next when a later node depends on an earlier one being reachable. Seereferences/operations.mdfor full lifecycle details.
What ships with it
5 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 5d ago Changed · +7 lines 4a8243306c57
- 2mo ago First seen · 107 lines · 141 tokens per session scan A c4fed07f4e30
cml is a skill published in the GitHub repository xorrkaz/cml-mcp (78 stars, last pushed 6d ago), licensed BSD-2-Clause. It adds 141 tokens to every session and 3,218 once invoked, about $0.0006 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
cml compared with skills from other repositories
| Mod | Type | Stars | Last push | Licence | Scan | Tokens | Works with |
|---|---|---|---|---|---|---|---|
| cmlxorrkaz/cml-mcp | Skill | 78 | 6d ago | BSD-2-Clause | A | 141 | Claude Code, Codex |
| opik-diagnosecomet-ml/opik-mcp | Skill | 226 | 3d ago | Apache-2.0 | A | 187 | Claude Code |
| ue-mcp-epic-routingdb-lyon/ue-mcp | Skill | 480 | 2d ago | MIT | A | 81 | Claude Code, Codex |
| qa-knowledgegriddynamics/rosetta | Skill | 373 | yesterday | Apache-2.0 | A | 32 | Claude Code, Codex |
| frontmcp-setupagentfront/frontmcp | Skill | 148 | yesterday | Apache-2.0 | A | 176 | Claude Code, Codex |
| apitapn1byn1kt/apitap | Skill | 131 | 19d ago | Apache-2.0 | A | 0 | Claude Code, Codex |
| x-use-reviewihuzaifashoukat/x-use | Skill | 181 | 7d ago | MIT | A | 56 | Claude Code |