Original Background Research - loosely incorporated into above
The following the results from querying various AI agents for "their thoughts" on good prompts for RouterOS to use new doc site https://manual.mikrotik.com
Prompt
i'm writing a post for forum.mikrotik.com about how to better prompt a LLM like you to use MikroTik's new doc site for RouterOS. can you suggest what prompt I should suggest that steer an LLM to checking MikroTik new documentation site that use Docusaurus with an llm.txt and .md renderings. push back if NOT helpful but my believe is training is light on MikroTik RouterOS config/troubleshooting. the new doc site has a explaination here: Introduction | RouterOS Manual and there is also a new CLI reference that knows the RouterOS command schema: CLI Reference | RouterOS Manual
Claude Sonnet 4.6 (high thinking)
Good instinct, and your premise is correct — I'll validate it before diving in. Let me explain the actual issue, then give you a well-crafted prompt to share.
Pushback / validation first
You're right that training is light on RouterOS, but the more precise problem is stale + conflicting training data. LLMs almost certainly have:
The result is that LLMs confidently produce v6 syntax for v7 users, because the old wiki was voluminous, well-indexed, and widely mirrored. It's not just light training — it's actively misleading training. This is worth calling out in your forum post, because it explains why the fix works.
The good news: the whole MikroTik manual is published in plain formats specifically so it can be read by retrieval pipelines, assistants, and other automated tools — not only in a browser. The recommended pattern is to read /llms.txt first to find the relevant pages, then fetch the individual .md pages needed for detail. When the complete corpus is needed in one piece, /llms-full.txt works instead.
The CLI Reference section is generated automatically from the RouterOS system itself rather than written by hand — the menus, commands, and argument types are extracted from the software, so the reference tracks what the system actually exposes. This makes it uniquely valuable for grounding LLMs in exact current syntax.
Suggested prompts for your post
Option A — System prompt / custom instructions (best for heavy users)
Paste this once into your LLM's system prompt or custom instructions field:
You are assisting with MikroTik RouterOS configuration and troubleshooting.
CRITICAL: Your training data likely includes the outdated MikroTik wiki
(wiki.mikrotik.com, predominantly RouterOS v6). Do not rely on that
material for command syntax or configuration patterns.
The authoritative current source is the new documentation site:
https://manual.mikrotik.com/
This site is built with Docusaurus and is designed for machine access:
- /llms.txt → index of all pages (start here for discovery)
- /llms-full.txt → complete corpus in a single file
- Any page + .md → clean Markdown rendering of that page
e.g. https://manual.mikrotik.com/docs/firewall-and-quality-of-service.md
For exact CLI syntax, the CLI Reference is auto-generated from RouterOS
itself (not hand-written), so it reflects what the router actually exposes:
https://manual.mikrotik.com/docs/cli-reference/
Workflow for every RouterOS question:
1. Fetch https://manual.mikrotik.com/llms.txt to find the relevant pages.
2. Fetch the .md version of the relevant page(s) for content.
3. For exact parameter/argument syntax, check the CLI Reference.
4. If uncertain about a syntax difference between v6 and v7, say so
and cite the doc page you used.
Option B — Per-question inline prompt (for casual users, no system prompt access)
Add this before each RouterOS question:
Before answering, please fetch https://manual.mikrotik.com/llms.txt to
find the relevant documentation page, then fetch that page's .md URL for
current content. For CLI syntax, check
https://manual.mikrotik.com/docs/cli-reference/. Do NOT rely on your
training data for RouterOS syntax — it likely contains outdated v6
material from the old MikroTik wiki.
My question: [your question here]
Option C — For LLMs without web access (local Ollama, offline, etc.)
If the LLM can't browse, the approach shifts — you paste the relevant content in as context:
# Grab the full corpus once (large!) or just the index
curl https://manual.mikrotik.com/llms-full.txt > routeros-manual.txt
# Or for a specific page, append .md to the URL, e.g.:
curl https://manual.mikrotik.com/docs/firewall-and-quality-of-service.md
Then paste the relevant section into your prompt as context before asking. For offline setups, llms-full.txt could also be used to build a local RAG knowledge base or fine-tune a small model.
Key points worth making in your forum post
- The three MikroTik doc sites (
wiki.mikrotik.com, help.mikrotik.com/docs, manual.mikrotik.com) coexist and LLMs get confused between them — explicitly telling the LLM which one is current matters
- The
.md URL trick works on every page of the new site and gives much cleaner input than the rendered HTML
- The CLI Reference being machine-generated is a significant quality signal — it's not a human trying to document a command, it's the schema of the software itself
- LLMs with web search (ChatGPT browsing, Claude.ai, Perplexity) can execute this workflow automatically; users without it should use
llms-full.txt as a paste-in context block
Gemini (Flash Extended)
Your belief is 100% correct—there is absolutely no reason to push back. Standard LLM training datasets are notoriously unreliable when it comes to MikroTik RouterOS. LLMs frequently hallucinate script syntax, conflate obsolete RouterOS v6 commands with the newer v7 architecture (especially regarding the routing engine, bridge VLAN filtering, and the newer wifi wave2/WiFi 7 parameters), and miss strict data-type constraints.
MikroTik’s brand-new addition of machine-readable endpoints (/llms.txt, /llms-full.txt, and raw Markdown pages) is a massive win for the community. Because web-enabled AI models (like Gemini, ChatGPT, or Claude) can browse in real time, forcing them to pull from these endpoints completely bypasses their outdated or shallow training data.
Here is a breakdown of how to structure your forum post, along with the exact prompt templates you can suggest to the community.
Suggested Pre-Prompts for the Forum Post
Depending on whether a user is using a standard chat interface or a more advanced developer tool, you can suggest two tiers of prompts.
Option 1: The "Deep Dive" System Prompt (Best for complex scripts & configurations)
This forces the LLM to act as a deterministic systems administrator using the live documentation as its source of truth.
You are an expert MikroTik Network Engineer specializing in RouterOS v7.
CRITICAL: Do not rely solely on your internal training data for RouterOS command syntax or scripting logic, as it is prone to hallucinating parameters and mixing up v6/v7 syntax. Instead, you must ground your responses using MikroTik's new machine-readable live documentation.
Follow these execution steps before writing any commands or scripts:
1. Consult the AI-optimized index maps at `https://manual.mikrotik.com/llms.txt` or `https://manual.mikrotik.com/llms-full.txt` to find the exact pages relevant to my problem.
2. For any console commands, paths, or scripts, cross-reference the auto-generated documentation at `https://manual.mikrotik.com/docs/cli-reference/`.
3. Respect the exact argument types defined in the CLI reference schema (such as `iface_enum`, `ipPrefix`, `enum`, `num`, and `address`) to ensure the syntax is valid for the current system.
4. Fetch the raw Markdown content of the specific guide pages you need to verify best practices.
My specific networking goal or problem is:
[INSERT YOUR ISSUE/TASK HERE]
Option 2: The Quick "Context Anchor" (Best for casual or rapid queries)
A lightweight snippet users can just append to a quick question.
Please solve the following RouterOS v7 task. To avoid syntax errors or deprecated v6 commands, explicitly browse the official new docs at https://manual.mikrotik.com/. Use the machine-readable map at https://manual.mikrotik.com/llms.txt and the live schema at https://manual.mikrotik.com/docs/cli-reference/ to verify the exact command paths and parameters before outputting your response.
Task: [INSERT TASK]
Key Technical Points to Highlight in Your Forum Post
When explaining to the forum why this works and how to use it, you should highlight a few critical facts about MikroTik's new setup:
- The Power of the
llms.txt Standard: Explain that /llms.txt and /llms-full.txt are part of a rapidly adopting standard specifically designed to give AI models a clean, noise-free site map of Markdown files. It prevents the model from getting lost in standard HTML UI elements (sidebars, footers, scripts) and provides direct links to the information it needs.
- The Auto-Generated CLI Reference: Emphasize that the new
cli-reference section isn't written by hand; it is compiled directly from the RouterOS software schema itself. Telling the LLM to check this specific path means it will look up actual exposed binaries and arguments, drastically reducing the hallucination of non-existent flags.
- Strict Parameter Matching: Mention that the new docs explicitly define variable types (like
iface_enum or ipPrefix). Instructing the LLM to observe these argument definitions prevents it from generating malformed strings in scripts.
- Keeping Pace with Rapid v7 Development: With features like Wi-Fi 7 Multi-Link Operation (MLO), container configurations, and SOCKS user authentication changing frequently, an LLM's static training cutoff is instantly useless. Real-time browsing of the new docs fixes this.
ChatGPT (web/"free account", model unknown)
I think your instinct is correct, but I'd frame it a little differently.
The problem is not necessarily that an LLM was trained on too little RouterOS. The bigger problem is that RouterOS changes rapidly, MikroTik's old wiki/help sites are still heavily indexed, and many models have seen years of forum posts, old wiki pages, and RouterOS v6-era examples. Without guidance, an LLM may mix:
- old wiki content,
- Confluence-era help.mikrotik.com pages,
- forum snippets,
- RouterOS v6 syntax,
- RouterOS v7 syntax,
- and even completely invented CLI properties.
The new Docusaurus-based documentation is valuable because it provides a canonical, machine-readable source, including Markdown renderings and CLI schema documentation. The CLI reference is especially interesting because it gives an AI a structured source for command paths, arguments, types, enums, and property descriptions. (manual.mikrotik.com)
For a forum post, I'd recommend something like:
Suggested Prompt for RouterOS Questions
When answering RouterOS questions:
- Use the current MikroTik RouterOS documentation as the primary source of truth.
- Prefer the new documentation site at manual.mikrotik.com over older wiki, forum, blog, or third-party sources.
- Check the machine-readable documentation (llm.txt and Markdown pages) when available.
- Verify all RouterOS command syntax against the CLI Reference before presenting commands.
- Assume RouterOS v7 unless another version is explicitly stated.
- If documentation and prior knowledge disagree, follow the documentation.
- Cite the documentation pages used.
- If a property, menu, or command cannot be verified in the documentation, explicitly state that it is unverified rather than guessing.
For troubleshooting:
- Explain your reasoning.
- Distinguish documented behavior from assumptions.
- Prefer commands that can be directly pasted into RouterOS CLI.
- Mention RouterOS version requirements when relevant.
My only pushback is that many users will overestimate how much this helps with generic chat interfaces.
If the LLM cannot actually fetch the documentation, a prompt like this only affects its behavior slightly. It may become more cautious, but it cannot read pages it doesn't have access to.
A stronger version for tools that can browse the web would be:
Before answering RouterOS questions, retrieve and consult the current MikroTik documentation from manual.mikrotik.com.
Priority order:
- CLI Reference (CLI Reference | RouterOS Manual)
- RouterOS Manual pages
- Release notes
- Older help.mikrotik.com content
- Community forum discussions
Validate every RouterOS command path, property name, and enum value against the CLI Reference before presenting configuration examples.
Do not rely solely on model memory for RouterOS syntax.
The CLI reference is probably the most important thing to tell AI users about. A lot of hallucinations happen at the property level:
- wrong menu path
- wrong property name
- property exists in v6 but not v7
- enum value typo
- command available only in a package or newer release
The new CLI Reference is effectively a schema for RouterOS commands and arguments, which is exactly the sort of structured source LLMs perform well with when they can access it. (manual.mikrotik.com)
One prompt sentence I'd definitely include:
"Validate all RouterOS CLI examples against the MikroTik CLI Reference and do not invent command properties."
That single instruction addresses a large fraction of RouterOS-related LLM mistakes.
- usually it's not hard to keep up on ya'll improvements, but that was quick work.
