@huseynsnmz/pi-suite 0.1.0 → 0.2.1

package/packages/web/src/markdown.ts

@@ -0,0 +1,156 @@
+/** Page HTML to Markdown. The page is hostile, so this only ever emits text, and escapes what would read as Markdown. */
+import { parseHTML } from "linkedom";
+
+interface DomNode {
+  nodeType: number;
+  tagName?: string;
+  textContent: string | null;
+  childNodes: Iterable<DomNode>;
+  getAttribute?(name: string): string | null;
+}
+
+const TEXT = 3;
+const ELEMENT = 1;
+const NEVER = new Set(["script", "style", "noscript", "template", "head", "title", "meta", "link", "iframe", "object", "embed", "svg", "canvas", "form", "input", "textarea", "select", "button", "img", "picture", "video", "audio"]);
+const BLOCK = new Set(["p", "div", "section", "article", "main", "aside", "header", "footer", "nav", "figure", "figcaption", "address", "details", "summary", "dl", "dt", "dd", "center"]);
+
+function escaped(text: string): string {
+  return text
+    .replace(/\\/g, "\\\\")
+    .replace(/[*`[\]_]/g, "\\$&")
+    .replace(/^(\s*)([-+=>]|#|~~~)/, "$1\\$2")
+    .replace(/^(\s*\d+)\./, "$1\\.");
+}
+
+/** Text with backticks in it needs a longer run of them around it. */
+function fenceFor(text: string, shortest: number): string {
+  let longest = 0;
+  for (const run of text.matchAll(/`+/g)) longest = Math.max(longest, run[0].length);
+  return "`".repeat(Math.max(shortest, longest + 1));
+}
+
+function inside(node: DomNode): string {
+  return [...node.childNodes].map(render).join("");
+}
+
+function list(node: DomNode, ordered: boolean): string {
+  let number = Number.parseInt(node.getAttribute?.("start") ?? "", 10);
+  if (!Number.isSafeInteger(number)) number = 1;
+  const items: string[] = [];
+  for (const child of node.childNodes) {
+    if (child.tagName?.toLowerCase() !== "li") {
+      // Not valid HTML, but browsers show it, so it's part of the page.
+      const stray = render(child).trim();
+      if (stray !== "") items.push(`  ${stray.replace(/\n/g, "\n  ")}`);
+      continue;
+    }
+    const marker = ordered ? `${number++}. ` : "- ";
+    const body = inside(child).trim().replace(/\n{3,}/g, "\n\n");
+    items.push(marker + body.replace(/\n/g, `\n${" ".repeat(marker.length)}`));
+  }
+  return `\n\n${items.join("\n")}\n\n`;
+}
+
+/** Marks go right around the words, and the spaces beside them stay outside, or the words run together. */
+function wrapped(text: string, wrap: (words: string) => string): string {
+  const words = text.trim();
+  if (words === "") return text;
+  const at = text.indexOf(words);
+  return text.slice(0, at) + wrap(words) + text.slice(at + words.length);
+}
+
+/** Only these can be followed from a page. The characters that could end a Markdown link are encoded. */
+const LINKABLE = /^(?:https?|mailto):/i;
+function linkTarget(href: string): string {
+  return href.replace(/[()\s]/g, (char) => `%${char.charCodeAt(0).toString(16).toUpperCase().padStart(2, "0")}`);
+}
+
+function render(node: DomNode): string {
+  if (node.nodeType === TEXT) return escaped((node.textContent ?? "").replace(/\s+/g, " "));
+  // Comments are never shown, so they don't count as the page's text.
+  if (node.nodeType !== ELEMENT) return "";
+  const tag = (node.tagName ?? "").toLowerCase();
+  if (NEVER.has(tag)) return "";
+  const heading = /^h([1-6])$/.exec(tag)?.[1];
+  if (heading !== undefined) {
+    const text = inside(node).trim();
+    return text === "" ? "" : `\n\n${"#".repeat(Number(heading))} ${text.replace(/\s*\n\s*/g, " ")}\n\n`;
+  }
+  if (BLOCK.has(tag)) return `\n\n${inside(node)}\n\n`;
+  switch (tag) {
+    case "br":
+      return "\n";
+    case "hr":
+      return "\n\n---\n\n";
+    case "ul":
+    case "ol":
+      return list(node, tag === "ol");
+    case "li":
+      return `\n- ${inside(node).trim()}\n`;
+    case "blockquote": {
+      const text = inside(node).trim().replace(/\n{3,}/g, "\n\n");
+      return text === "" ? "" : `\n\n${text.replace(/^/gm, "> ")}\n\n`;
+    }
+    case "pre": {
+      const text = (node.textContent ?? "").replace(/\n$/, "");
+      const fence = fenceFor(text, 3);
+      const code = [...node.childNodes].find((child) => child.tagName?.toLowerCase() === "code");
+      const language = /language-([\w+-]+)/.exec(code?.getAttribute?.("class") ?? "")?.[1] ?? "";
+      return `\n\n${fence}${language}\n${text}\n${fence}\n\n`;
+    }
+    case "code": {
+      const text = (node.textContent ?? "").replace(/\s+/g, " ");
+      if (text.trim() === "") return "";
+      const fence = fenceFor(text, 1);
+      const pad = text.startsWith("`") || text.endsWith("`") ? " " : "";
+      return `${fence}${pad}${text}${pad}${fence}`;
+    }
+    case "strong":
+    case "b":
+    case "em":
+    case "i": {
+      const mark = tag === "strong" || tag === "b" ? "**" : "_";
+      return wrapped(inside(node), (words) => `${mark}${words}${mark}`);
+    }
+    case "a": {
+      const text = inside(node).replace(/\n+/g, " ");
+      const href = node.getAttribute?.("href");
+      if (!href || !LINKABLE.test(href)) return text;
+      return wrapped(text, (words) => `[${words}](${linkTarget(href)})`);
+    }
+    case "table":
+      return `\n\n${[...node.childNodes].map(render).join("").trim()}\n\n`;
+    case "thead":
+    case "tbody":
+    case "tfoot":
+      return [...node.childNodes].filter((child) => child.nodeType === ELEMENT).map(render).join("");
+    case "tr":
+      return `${[...node.childNodes].map((cell) => render(cell).trim()).filter(Boolean).join(" | ")}\n`;
+    default:
+      return inside(node);
+  }
+}
+
+/** Removes trailing spaces and extra blank lines, a line at a time so a page of spaces costs nothing, and leaves code blocks exactly as they are. */
+function tidied(markdown: string): string {
+  const out: string[] = [];
+  let fence: string | undefined;
+  for (const line of markdown.split("\n")) {
+    if (fence !== undefined) {
+      out.push(line);
+      if (line === fence) fence = undefined;
+      continue;
+    }
+    // Page text has its backticks escaped, so only a real code block can start one of these.
+    const opening = /^(`{3,})[\w+-]*$/.exec(line)?.[1];
+    if (opening !== undefined) fence = opening;
+    const tidy = line.trimEnd();
+    if (tidy !== "" || (out.length > 0 && out.at(-1) !== "")) out.push(tidy);
+  }
+  return out.join("\n").trim();
+}
+
+export function markdownOf(html: string): string {
+  const { document } = parseHTML(`<!doctype html><html><body>${html}</body></html>`);
+  return tidied(render(document.body as unknown as DomNode));
+}

package/src/index.ts

@@ -1,4 +1,4 @@
-// One package, six parts. A part that's off isn't even imported, so it adds no tools, prompts or start-up time.
+// One package, seven parts. A part that's off isn't even imported, so it adds no tools, prompts or start-up time.
 import { readFileSync } from "node:fs";
 import { join } from "node:path";
 import { fileURLToPath } from "node:url";

package/src/settings.ts

@@ -9,6 +9,7 @@ export const PARTS = {
   memory: "Memory the agent keeps between sessions",
   frame: "The start card, framed editor, tool boxes, footer and todo list",
   board: "The board of agent sessions",
+  commands: "/clear and /security-review",
 } as const;
 
 export type Part = keyof typeof PARTS;