roblox-docs

Roblox creator-docs search

This skill answers Roblox/Luau questions from the official docs instead of memory. It works off a local, auto-updating clone of Roblox/creator-docs, so it stays current and can full-text search every class, datatype, enum, global, library, and guide at once — something live web fetching of single doc pages can't do.

When to reach for this

This is primarily a tool for you, the agent, to keep yourself honest about an API surface that changes faster than training data. Don't wait for the user to ask for "docs". Use it when:

• You're unsure. You're about to write or recommend an API call (a method name, its arguments, return type, an enum value, a property) and you aren't fully confident it's correct and current. Verifying costs little; a hallucinated signature costs the user real debugging time.
• You need depth. The task needs more than a name — behavior, edge cases, deprecations, replacements, or how pieces fit together.
• You're exploring alternatives. There's more than one API/approach for the goal and you want to compare what the docs actually recommend.
• You're planning. Especially during planning, before you lock in an approach — check the docs first so the plan isn't built on wrong assumptions. A plan grounded in real APIs beats one you have to unwind later.

When in doubt, look it up. The whole point is to replace confident guessing with cited fact.

Why a subagent

Doc files are large (a single class YAML can be thousands of lines). Reading several of them directly into this conversation would bury the actual answer in raw reference text. So delegate the lookup to an Explore subagent: it does the searching and full-file reading in its own context and returns only a tight, cited answer. Your context stays clean.

How to use it

When a Roblox/Luau question comes up, dispatch one Explore subagent (via the Agent tool) with the prompt below. The search script lives at:

${CLAUDE_PLUGIN_ROOT}/skills/roblox-docs/scripts/robloxdocs.py

Substitute that absolute path and the user's question into this prompt:

You are looking up official Roblox documentation to answer a question. Use the bundled search tool — do not fetch the web and do not answer from memory.

1. Run: python3 <SCRIPT_PATH> "<search query>" The first run clones the docs (~1 min); later runs reuse a cached clone and auto-pull when stale. It prints ranked path:line matches with snippets.
2. Pick the most relevant files and Read them in full (the snippets are only for ranking — the real answer is in the file).
3. If the first query misses, try again with different terms (an exact API name, a related class, or the feature's plain-English name). The tool accepts dotted/colon forms like TweenService:Create.
4. Return a concise answer: the relevant API signatures / steps, a short code example if useful, and cite each fact as path:line from the clone. If the docs genuinely don't cover it, say so — don't invent API.

Question: <THE USER'S QUESTION>

Then relay the subagent's answer to the user, keeping its citations.

Notes

• Prerequisites: git and Python 3 on PATH. ripgrep (rg) is used if present for faster search but isn't required.
• Freshness: the clone auto-pulls when older than 24h. To force a refresh, the subagent (or you) can run python3 <SCRIPT_PATH> --update.
• Multiple questions: batch related lookups into one subagent prompt so it can search and read once rather than spinning up repeatedly.
• Offline: if a pull fails, the tool falls back to the cached clone and notes it — the answer may be slightly stale but still works.