diff --git a/.github/workflows/skill-generator.yml b/.github/workflows/skill-generator.yml
new file mode 100644
index 0000000..7af6694
--- /dev/null
+++ b/.github/workflows/skill-generator.yml
@@ -0,0 +1,48 @@
+name: sync-sandbox-agent-skill
+
+on:
+  push:
+    branches: [main]
+  workflow_dispatch:
+
+jobs:
+  generate:
+    runs-on: ubuntu-24.04
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: actions/setup-node@v4
+        with:
+          node-version: 20
+
+      - name: Generate skill artifacts
+        run: node scripts/skill-generator/generate.js
+
+      - name: Sync to skills repo
+        env:
+          SKILLS_REPO_TOKEN: ${{ secrets.SKILLS_REPO_TOKEN }}
+        run: |
+          if [ -z "$SKILLS_REPO_TOKEN" ]; then
+            echo "SKILLS_REPO_TOKEN is not set" >&2
+            exit 1
+          fi
+
+          git config --global user.name "github-actions[bot]"
+          git config --global user.email "github-actions[bot]@users.noreply.github.com"
+
+          git clone "https://x-access-token:${SKILLS_REPO_TOKEN}@github.com/rivet-dev/skills.git" /tmp/rivet-skills
+
+          mkdir -p /tmp/rivet-skills/skills/sandbox-agent
+          rm -rf /tmp/rivet-skills/skills/sandbox-agent/*
+          cp -R scripts/skill-generator/dist/* /tmp/rivet-skills/skills/sandbox-agent/
+
+          cd /tmp/rivet-skills
+          git add skills/sandbox-agent
+
+          if git diff --cached --quiet; then
+            echo "No skill changes to publish"
+            exit 0
+          fi
+
+          git commit -m "chore: update sandbox-agent skill"
+          git push
diff --git a/scripts/skill-generator/generate.js b/scripts/skill-generator/generate.js
new file mode 100644
index 0000000..37fbc8c
--- /dev/null
+++ b/scripts/skill-generator/generate.js
@@ -0,0 +1,364 @@
+#!/usr/bin/env node
+
+const fs = require("node:fs");
+const fsp = require("node:fs/promises");
+const path = require("node:path");
+
+const DOCS_ROOT = path.resolve(__dirname, "..", "..", "docs");
+const OUTPUT_ROOT = path.resolve(__dirname, "dist");
+const TEMPLATE_PATH = path.resolve(__dirname, "template", "SKILL.md");
+const DOCS_BASE_URL = "https://sandboxagent.dev/docs";
+
+async function main() {
+  if (!fs.existsSync(DOCS_ROOT)) {
+    throw new Error(`Docs directory not found at ${DOCS_ROOT}`);
+  }
+
+  await fsp.rm(OUTPUT_ROOT, { recursive: true, force: true });
+  await fsp.mkdir(path.join(OUTPUT_ROOT, "reference"), { recursive: true });
+
+  const docFiles = await listDocFiles(DOCS_ROOT);
+  const references = [];
+
+  for (const filePath of docFiles) {
+    const relPath = normalizePath(path.relative(DOCS_ROOT, filePath));
+    const raw = await fsp.readFile(filePath, "utf8");
+    const { data, body } = parseFrontmatter(raw);
+
+    const slug = toSlug(relPath);
+    const canonicalUrl = slug ? `${DOCS_BASE_URL}/${slug}` : DOCS_BASE_URL;
+    const title = data.title || titleFromSlug(slug || relPath);
+    const description = data.description || "";
+
+    const markdown = convertDocToMarkdown(body);
+
+    const referenceRelPath = `${stripExtension(relPath)}.md`;
+    const outputPath = path.join(OUTPUT_ROOT, "reference", referenceRelPath);
+    await fsp.mkdir(path.dirname(outputPath), { recursive: true });
+
+    const referenceFile = buildReferenceFile({
+      title,
+      description,
+      canonicalUrl,
+      sourcePath: `docs/${relPath}`,
+      body: markdown,
+    });
+
+    await fsp.writeFile(outputPath, referenceFile, "utf8");
+
+    references.push({
+      slug,
+      title,
+      description,
+      canonicalUrl,
+      referencePath: `reference/${referenceRelPath}`,
+    });
+  }
+
+  const quickstart = references.find((ref) => ref.slug === "quickstart");
+  if (!quickstart) {
+    throw new Error("Quickstart doc not found. Expected docs/quickstart.mdx");
+  }
+
+  const quickstartPath = path.join(DOCS_ROOT, "quickstart.mdx");
+  const quickstartRaw = await fsp.readFile(quickstartPath, "utf8");
+  const { body: quickstartBody } = parseFrontmatter(quickstartRaw);
+  const quickstartContent = convertDocToMarkdown(quickstartBody);
+
+  const referenceMap = buildReferenceMap(references);
+  const template = await fsp.readFile(TEMPLATE_PATH, "utf8");
+
+  const skillFile = template
+    .replace("{{QUICKSTART}}", quickstartContent)
+    .replace("{{REFERENCE_MAP}}", referenceMap);
+
+  await fsp.writeFile(path.join(OUTPUT_ROOT, "SKILL.md"), `${skillFile.trim()}
+`, "utf8");
+
+  console.log(`Generated skill files in ${OUTPUT_ROOT}`);
+}
+
+async function listDocFiles(dir) {
+  const entries = await fsp.readdir(dir, { withFileTypes: true });
+  const files = [];
+
+  for (const entry of entries) {
+    const fullPath = path.join(dir, entry.name);
+    if (entry.isDirectory()) {
+      files.push(...(await listDocFiles(fullPath)));
+      continue;
+    }
+    if (!entry.isFile()) continue;
+    if (!/\.mdx?$/.test(entry.name)) continue;
+    files.push(fullPath);
+  }
+
+  return files;
+}
+
+function parseFrontmatter(content) {
+  if (!content.startsWith("---")) {
+    return { data: {}, body: content.trim() };
+  }
+
+  const match = content.match(/^---\n([\s\S]*?)\n---\n?/);
+  if (!match) {
+    return { data: {}, body: content.trim() };
+  }
+
+  const frontmatter = match[1];
+  const body = content.slice(match[0].length);
+  const data = {};
+
+  for (const line of frontmatter.split("\n")) {
+    const trimmed = line.trim();
+    if (!trimmed || trimmed.startsWith("#")) continue;
+    const idx = trimmed.indexOf(":");
+    if (idx === -1) continue;
+    const key = trimmed.slice(0, idx).trim();
+    let value = trimmed.slice(idx + 1).trim();
+    value = value.replace(/^"|"$/g, "").replace(/^'|'$/g, "");
+    data[key] = value;
+  }
+
+  return { data, body: body.trim() };
+}
+
+function toSlug(relPath) {
+  const withoutExt = stripExtension(relPath);
+  const normalized = withoutExt.replace(/\\/g, "/");
+  if (normalized.endsWith("/index")) {
+    return normalized.slice(0, -"/index".length);
+  }
+  return normalized;
+}
+
+function stripExtension(value) {
+  return value.replace(/\.mdx?$/i, "");
+}
+
+function titleFromSlug(value) {
+  const cleaned = value.replace(/\.mdx?$/i, "").replace(/\\/g, "/");
+  const parts = cleaned.split("/").filter(Boolean);
+  const last = parts[parts.length - 1] || "index";
+  return formatSegment(last);
+}
+
+function buildReferenceFile({ title, description, canonicalUrl, sourcePath, body }) {
+  const lines = [
+    `# ${title}`,
+    "",
+    `> Source: \`${sourcePath}\``,
+    `> Canonical URL: ${canonicalUrl}`,
+    `> Description: ${description || ""}`,
+    "",
+    "---",
+    body.trim(),
+  ];
+
+  return `${lines.join("\n").trim()}\n`;
+}
+
+function buildReferenceMap(references) {
+  const grouped = new Map();
+  const groupRoots = new Set();
+
+  for (const ref of references) {
+    const segments = (ref.slug || "").split("/").filter(Boolean);
+    if (segments.length > 1) {
+      groupRoots.add(segments[0]);
+    }
+  }
+
+  for (const ref of references) {
+    const segments = (ref.slug || "").split("/").filter(Boolean);
+    let group = "general";
+    if (segments.length > 1) {
+      group = segments[0];
+    } else if (segments.length === 1 && groupRoots.has(segments[0])) {
+      group = segments[0];
+    }
+
+    if (!grouped.has(group)) grouped.set(group, []);
+    grouped.get(group).push(ref);
+  }
+
+  const lines = [];
+  const sortedGroups = [...grouped.keys()].sort((a, b) => a.localeCompare(b));
+
+  for (const group of sortedGroups) {
+    lines.push(`### ${formatSegment(group)}`, "");
+    const items = grouped.get(group).slice().sort((a, b) => a.title.localeCompare(b.title));
+    for (const item of items) {
+      lines.push(`- [${item.title}](${item.referencePath})`);
+    }
+    lines.push("");
+  }
+
+  return lines.join("\n").trim();
+}
+
+function formatSegment(value) {
+  if (!value) return "General";
+  const special = {
+    ai: "AI",
+    sdks: "SDKs",
+  };
+  if (special[value]) return special[value];
+  if (value === "general") return "General";
+  return value
+    .split("-")
+    .map((part) => part.charAt(0).toUpperCase() + part.slice(1))
+    .join(" ");
+}
+
+function normalizePath(value) {
+  return value.replace(/\\/g, "/");
+}
+
+function convertDocToMarkdown(body) {
+  const { replaced, restore } = extractCodeBlocks(body ?? "");
+  let text = replaced;
+
+  text = text.replace(/^[ \t]*import\s+[^;]+;?\s*$/gm, "");
+  text = text.replace(/^[ \t]*export\s+[^;]+;?\s*$/gm, "");
+  text = text.replace(/\{\/\*[\s\S]*?\*\/\}/g, "");
+
+  text = stripWrapperTags(text, "Steps");
+  text = stripWrapperTags(text, "Tabs");
+  text = stripWrapperTags(text, "CardGroup");
+  text = stripWrapperTags(text, "CodeGroup");
+  text = stripWrapperTags(text, "AccordionGroup");
+  text = stripWrapperTags(text, "Frame");
+
+  text = formatHeadingBlocks(text, "Step", "Step", 3);
+  text = formatHeadingBlocks(text, "Tab", "Tab", 4);
+  text = formatHeadingBlocks(text, "Accordion", "Details", 4);
+
+  text = formatCards(text);
+
+  text = applyCallouts(text, "Tip");
+  text = applyCallouts(text, "Note");
+  text = applyCallouts(text, "Warning");
+  text = applyCallouts(text, "Info");
+  text = applyCallouts(text, "Callout");
+
+  text = replaceImages(text);
+
+  text = text.replace(/<Card[^>]*>/gi, "").replace(/<\/Card>/gi, "");
+  text = text.replace(/<Steps[^>]*>/gi, "").replace(/<\/Steps>/gi, "");
+  text = text.replace(/<Tabs[^>]*>/gi, "").replace(/<\/Tabs>/gi, "");
+  text = text.replace(/<Step[^>]*>/gi, "").replace(/<\/Step>/gi, "");
+  text = text.replace(/<Tab[^>]*>/gi, "").replace(/<\/Tab>/gi, "");
+  text = text.replace(/<Accordion[^>]*>/gi, "").replace(/<\/Accordion>/gi, "");
+  text = text.replace(/<Frame[^>]*>/gi, "").replace(/<\/Frame>/gi, "");
+
+  text = text.replace(/<[A-Z][A-Za-z0-9]*[^>]*>/g, "").replace(/<\/[A-Z][A-Za-z0-9]*>/g, "");
+  text = stripIndentation(text);
+  text = text.replace(/\n{3,}/g, "\n\n");
+
+  return restore(text).trim();
+}
+
+function extractCodeBlocks(input) {
+  const blocks = [];
+  const replaced = input.replace(/```[\s\S]*?```/g, (match) => {
+    const token = `@@CODE_BLOCK_${blocks.length}@@`;
+    blocks.push(normalizeCodeBlock(match));
+    return token;
+  });
+
+  return {
+    replaced,
+    restore: (value) => value.replace(/@@CODE_BLOCK_(\d+)@@/g, (_, index) => blocks[Number(index)] ?? ""),
+  };
+}
+
+function normalizeCodeBlock(block) {
+  const lines = block.split("\n");
+  if (lines.length < 2) return block.trim();
+
+  const opening = lines[0].trim();
+  const closing = lines[lines.length - 1].trim();
+  const contentLines = lines.slice(1, -1);
+  const indents = contentLines
+    .filter((line) => line.trim() !== "")
+    .map((line) => line.match(/^\s*/)?.[0].length ?? 0);
+  const minIndent = indents.length ? Math.min(...indents) : 0;
+  const normalizedContent = contentLines.map((line) => line.slice(minIndent));
+
+  return [opening, ...normalizedContent, closing].join("\n");
+}
+
+function stripWrapperTags(input, tag) {
+  const open = new RegExp(`<${tag}[^>]*>`, "gi");
+  const close = new RegExp(`</${tag}>`, "gi");
+  return input.replace(open, "\n").replace(close, "\n");
+}
+
+function formatHeadingBlocks(input, tag, fallback, level) {
+  const heading = "#".repeat(level);
+  const withTitles = input.replace(
+    new RegExp(`<${tag}[^>]*title=(?:\"([^\"]+)\"|'([^']+)')[^>]*>`, "gi"),
+    (_, doubleQuoted, singleQuoted) => `\n${heading} ${(doubleQuoted ?? singleQuoted ?? fallback).trim()}\n\n`,
+  );
+  const withFallback = withTitles.replace(new RegExp(`<${tag}[^>]*>`, "gi"), `\n${heading} ${fallback}\n\n`);
+  return withFallback.replace(new RegExp(`</${tag}>`, "gi"), "\n");
+}
+
+function formatCards(input) {
+  return input.replace(/<Card([^>]*)>([\s\S]*?)<\/Card>/gi, (_, attrs, content) => {
+    const title = getAttributeValue(attrs, "title") ?? "Resource";
+    const href = getAttributeValue(attrs, "href");
+    const summary = collapseWhitespace(stripHtml(content));
+    const link = href ? `[${title}](${href})` : title;
+    const suffix = summary ? ` — ${summary}` : "";
+    return `\n- ${link}${suffix}\n\n`;
+  });
+}
+
+function applyCallouts(input, tag) {
+  const regex = new RegExp(`<${tag}[^>]*>([\s\S]*?)</${tag}>`, "gi");
+  return input.replace(regex, (_, content) => {
+    const label = tag.toUpperCase();
+    const text = collapseWhitespace(stripHtml(content));
+    return `\n> **${label}:** ${text}\n\n`;
+  });
+}
+
+function replaceImages(input) {
+  return input.replace(/<img\s+([^>]+?)\s*\/?>(?:\s*<\/img>)?/gi, (_, attrs) => {
+    const src = getAttributeValue(attrs, "src") ?? "";
+    const alt = getAttributeValue(attrs, "alt") ?? "";
+    if (!src) return "";
+    const url = src.startsWith("/") ? `${DOCS_BASE_URL}${src}` : src;
+    return `![${alt}](${url})`;
+  });
+}
+
+function getAttributeValue(attrs, name) {
+  const regex = new RegExp(`${name}=(?:\"([^\"]+)\"|'([^']+)')`, "i");
+  const match = attrs.match(regex);
+  if (!match) return undefined;
+  return (match[1] ?? match[2] ?? "").trim();
+}
+
+function stripHtml(value) {
+  return value.replace(/<[^>]+>/g, " ").replace(/\s+/g, " ").trim();
+}
+
+function collapseWhitespace(value) {
+  return value.replace(/\s+/g, " ").trim();
+}
+
+function stripIndentation(input) {
+  return input
+    .split("\n")
+    .map((line) => line.replace(/^\t+/, "").replace(/^ {2,}/, ""))
+    .join("\n");
+}
+
+main().catch((error) => {
+  console.error(error);
+  process.exit(1);
+});
diff --git a/scripts/skill-generator/template/SKILL.md b/scripts/skill-generator/template/SKILL.md
new file mode 100644
index 0000000..629b494
--- /dev/null
+++ b/scripts/skill-generator/template/SKILL.md
@@ -0,0 +1,16 @@
+---
+name: "sandbox-agent"
+description: "Documentation and API reference for Sandbox Agent."
+---
+
+# Sandbox Agent
+
+Use this skill to deploy, configure, and integrate Sandbox Agent. Prefer citing canonical docs URLs from the reference map.
+
+## Quickstart
+
+{{QUICKSTART}}
+
+## Reference Map
+
+{{REFERENCE_MAP}}