dsh-aris-panel 0.1.0

package/skills/render-html/scripts/render_html.py

@@ -0,0 +1,1006 @@
+#!/usr/bin/env python3
+"""render_html.py — convert ARIS Markdown / JSON artifacts to single-file HTML.
+
+Pure-stdlib Python. No external pip deps.
+
+Usage:
+    render_html.py <input.md> [--template academic|dashboard]
+                              [--out <output.html>]
+                              [--title "..."] [--subtitle "..."]
+                              [--eyebrow "..."]
+                              [--state <state.json>]
+                              [--json <sidecar.json>]
+                              [--offline]
+                              [--no-toc]
+
+See skills/render-html/SKILL.md for the full contract.
+
+Design invariants (see codex review consultation in commit message):
+  - Markdown / JSON is canonical source, HTML is generated view.
+  - HTML embeds source path + SHA256 + generated timestamp (drift detection).
+  - Single-file output. MathJax + highlight.js loaded from CDN unless --offline.
+  - Pure stdlib: re, html, hashlib, json, datetime, pathlib, argparse, sys.
+  - Conservative Markdown subset matching what ARIS artifacts actually emit.
+"""
+from __future__ import annotations
+
+import argparse
+import hashlib
+import html as html_lib
+import json
+import re
+import sys
+from datetime import datetime, timezone
+from pathlib import Path
+
+TEMPLATES_DIR = Path(__file__).resolve().parent / "templates"
+
+# ---------------------------------------------------------------------------
+# Inline parsing
+# ---------------------------------------------------------------------------
+
+# Placeholders used to protect inline content during multi-pass rewriting.
+# Two non-printable PUA chars + index to avoid collisions with real content.
+_PH_OPEN = ""
+_PH_CLOSE = ""
+
+
+def _ph(idx: int) -> str:
+    return f"{_PH_OPEN}{idx}{_PH_CLOSE}"
+
+
+_RE_CODE_INLINE = re.compile(r"`([^`\n]+)`")
+_RE_MATH_DISPLAY = re.compile(r"\$\$([^\n][\s\S]*?)\$\$")
+_RE_MATH_INLINE = re.compile(r"(?<!\\)\$([^\$\n]+?)\$")
+_RE_IMG = re.compile(r"!\[([^\]]*)\]\(([^)\s]+)(?:\s+\"([^\"]*)\")?\)")
+_RE_LINK = re.compile(r"\[([^\]]+)\]\(([^)\s]+)(?:\s+\"([^\"]*)\")?\)")
+_RE_BOLD = re.compile(r"\*\*([^\*\n]+)\*\*")
+_RE_ITALIC = re.compile(r"(?<!\*)\*([^\*\n]+)\*(?!\*)")
+_RE_ITALIC_UNDERSCORE = re.compile(r"(?<!\w)_([^_\n]+)_(?!\w)")
+_RE_STRIKE = re.compile(r"~~([^~\n]+)~~")
+# Wikilink paper reference: [[key]] or [[key|display]]. Backslash-escaped form
+# `\[[...]]` is left alone so authors can opt out.
+_RE_PAPER_REF = re.compile(
+    r"(?<!\\)\[\[([A-Za-z0-9][A-Za-z0-9_.:-]*)(?:\|([^\]\n]+))?\]\]"
+)
+
+# Inline HTML tags that should pass through inline (commonly used in ARIS docs).
+_INLINE_HTML_TAGS = ("br", "img", "a", "span", "sub", "sup", "code", "kbd", "b", "i", "u", "strong", "em")
+
+# URL schemes considered safe for href/src. javascript:, data:, vbscript: blocked.
+_SAFE_URL_SCHEMES = ("http:", "https:", "mailto:", "ftp:", "tel:", "#", "/", "./", "../")
+
+
+def _safe_url(url: str) -> str:
+    """Return url if scheme is safe, else '#'. Defensive against javascript: links."""
+    s = url.strip().lower()
+    # Relative paths and fragments are fine
+    if s.startswith(("#", "/", "./", "../")) or s == "":
+        return url
+    if s.startswith(_SAFE_URL_SCHEMES):
+        return url
+    # Bare path like "foo/bar.md" or "page.html" — no scheme prefix, safe.
+    if ":" not in s.split("/", 1)[0]:
+        return url
+    # Block javascript:, data:, vbscript:, file:, etc.
+    return "#blocked-unsafe-url"
+
+
+# Tags stripped wholesale from HTML passthrough (block and inline).
+# Even if the workflow LLM hallucinates these, they never reach output.
+_RE_STRIP_TAG = re.compile(
+    r"<\s*(script|style|iframe|object|embed|form|input|button|link|meta|base)\b[^>]*>.*?</\s*\1\s*>",
+    re.IGNORECASE | re.DOTALL,
+)
+_RE_STRIP_TAG_SELF = re.compile(
+    r"<\s*(script|style|iframe|object|embed|form|input|button|link|meta|base)\b[^>]*/?\s*>",
+    re.IGNORECASE,
+)
+# Event-handler attributes like onclick=, onload=, etc. — strip these.
+_RE_STRIP_EVENT_ATTR = re.compile(
+    r"\s+on[a-z]+\s*=\s*(\"[^\"]*\"|'[^']*'|[^\s>]+)",
+    re.IGNORECASE,
+)
+# javascript:/vbscript:/data: in href/src — replace with #blocked.
+_RE_STRIP_DANGEROUS_URL_ATTR = re.compile(
+    r"""(\b(?:href|src|action|formaction|poster)\s*=\s*["']?)\s*(?:javascript|vbscript|data)\s*:""",
+    re.IGNORECASE,
+)
+
+
+def sanitize_html(s: str) -> str:
+    """Strip dangerous tags / event handlers / javascript: URLs from raw HTML.
+
+    Applied to: (a) inline-HTML spans we pass through, (b) block-HTML
+    passthrough content. Markdown text content is HTML-escaped separately
+    and never reaches this function. ARIS workflow artifacts should not
+    contain these tags; this is defense-in-depth in case an LLM hallucinates
+    one.
+    """
+    s = _RE_STRIP_TAG.sub("", s)
+    s = _RE_STRIP_TAG_SELF.sub("", s)
+    s = _RE_STRIP_EVENT_ATTR.sub("", s)
+    s = _RE_STRIP_DANGEROUS_URL_ATTR.sub(r"\1#blocked-unsafe-url:", s)
+    return s
+
+
+def render_inline(text: str) -> str:
+    """Convert Markdown inline syntax to HTML. Escape everything else.
+
+    Order matters:
+      1. Stash inline code, math (display+inline), and raw HTML tag-shaped
+         spans into placeholders (so we don't expand markdown inside them).
+      2. HTML-escape the remaining text.
+      3. Apply images, links, bold, italic, strike.
+      4. Restore placeholders.
+    """
+    stash: list[str] = []
+
+    def store(replacement: str) -> str:
+        idx = len(stash)
+        stash.append(replacement)
+        return _ph(idx)
+
+    # 1a. Inline code -- escape inner content fully.
+    def _code_sub(m: re.Match[str]) -> str:
+        return store(f"<code>{html_lib.escape(m.group(1))}</code>")
+
+    text = _RE_CODE_INLINE.sub(_code_sub, text)
+
+    # HTML-escape <, >, & inside math bodies: a bare "<t" (e.g. y_{<t}) is
+    # otherwise parsed as an HTML start tag and eats the rest of the formula
+    # before MathJax runs. MathJax v3 reads the decoded DOM text, so the escape
+    # is transparent to the TeX it sees (and & for cases/align stays intact).
+    # 1b. Display math (passthrough; MathJax will render).
+    def _md_sub(m: re.Match[str]) -> str:
+        body = html_lib.escape(m.group(1), quote=False)
+        return store(f"$${body}$$")
+
+    text = _RE_MATH_DISPLAY.sub(_md_sub, text)
+
+    # 1c. Inline math (passthrough).
+    def _mi_sub(m: re.Match[str]) -> str:
+        body = html_lib.escape(m.group(1), quote=False)
+        return store(f"${body}$")
+
+    text = _RE_MATH_INLINE.sub(_mi_sub, text)
+
+    # 1d. Wikilink paper refs [[key]] or [[key|display]] — stash as a clickable
+    # span with data-ref="key". The template JS wires up the popover from
+    # window.PAPER_REGISTRY (sidecar JSON loaded via --papers). Default display
+    # is the key uppercased; explicit display via |label form.
+    def _ref_sub(m: re.Match[str]) -> str:
+        key = m.group(1)
+        display = m.group(2) if m.group(2) is not None else key.upper()
+        return store(
+            f'<span data-ref="{html_lib.escape(key, quote=True)}">{html_lib.escape(display)}</span>'
+        )
+
+    text = _RE_PAPER_REF.sub(_ref_sub, text)
+
+    # 1e. Inline HTML spans (very limited allowlist + sanitize before stash).
+    _re_tag = re.compile(
+        r"<(/?)(" + "|".join(_INLINE_HTML_TAGS) + r")(\s[^<>]*)?>",
+        re.IGNORECASE,
+    )
+
+    def _tag_sub(m: re.Match[str]) -> str:
+        return store(sanitize_html(m.group(0)))
+
+    text = _re_tag.sub(_tag_sub, text)
+
+    # 2. HTML-escape remainder.
+    text = html_lib.escape(text, quote=False)
+
+    # 3. Apply markdown emphasis & links.
+    def _img_sub(m: re.Match[str]) -> str:
+        alt = html_lib.escape(m.group(1), quote=True)
+        src = html_lib.escape(_safe_url(m.group(2)), quote=True)
+        title = m.group(3)
+        title_attr = f' title="{html_lib.escape(title, quote=True)}"' if title else ""
+        return f'<img src="{src}" alt="{alt}"{title_attr} />'
+
+    text = _RE_IMG.sub(_img_sub, text)
+
+    def _link_sub(m: re.Match[str]) -> str:
+        label = m.group(1)  # inner label can still contain code placeholders; ok.
+        href = html_lib.escape(_safe_url(m.group(2)), quote=True)
+        title = m.group(3)
+        title_attr = f' title="{html_lib.escape(title, quote=True)}"' if title else ""
+        return f'<a href="{href}"{title_attr}>{label}</a>'
+
+    text = _RE_LINK.sub(_link_sub, text)
+    text = _RE_BOLD.sub(r"<strong>\1</strong>", text)
+    text = _RE_ITALIC.sub(r"<em>\1</em>", text)
+    text = _RE_ITALIC_UNDERSCORE.sub(r"<em>\1</em>", text)
+    text = _RE_STRIKE.sub(r"<del>\1</del>", text)
+
+    # 4. Restore placeholders.
+    def _restore(m: re.Match[str]) -> str:
+        idx = int(m.group(1))
+        return stash[idx]
+
+    text = re.sub(rf"{_PH_OPEN}(\d+){_PH_CLOSE}", _restore, text)
+    return text
+
+
+# ---------------------------------------------------------------------------
+# Block parsing
+# ---------------------------------------------------------------------------
+
+_RE_HEADING = re.compile(r"^(#{1,6})\s+(.+?)\s*#*\s*$")
+_RE_HR = re.compile(r"^\s*(?:-{3,}|\*{3,}|_{3,})\s*$")
+_RE_CODE_FENCE = re.compile(
+    r"^```\s*(?:(?P<lang>[A-Za-z0-9_+.#-]+)\s*)?(?:\{(?P<flags>[^}\n]+)\}\s*)?$"
+)
+_RE_TABLE_DIVIDER = re.compile(r"^\s*\|?\s*:?-+:?\s*(\|\s*:?-+:?\s*)+\|?\s*$")
+_RE_ORDERED = re.compile(r"^(\s*)(\d+)[.)]\s+(.*)$")
+_RE_UNORDERED = re.compile(r"^(\s*)[-*+]\s+(.*)$")
+_RE_BLOCKQUOTE = re.compile(r"^>\s?(.*)$")
+_RE_HTML_BLOCK_OPEN = re.compile(
+    r"^\s*<(details|div|figure|table|p|ul|ol|nav|section|aside|header|footer|main|article|blockquote)(\s|>|/>)",
+    re.IGNORECASE,
+)
+
+_CALLOUT_PREFIX_MAP = [
+    # (regex on first content, css class, default title)
+    (re.compile(r"^[⚠️⚠]️?\s*"), "callout-warn", "Warning"),
+    (re.compile(r"^💡\s*"), "callout-info", "Tip"),
+    (re.compile(r"^✅\s*"), "callout-good", "OK"),
+    (re.compile(r"^✓\s*"), "callout-good", "OK"),
+    (re.compile(r"^❌\s*"), "callout-bad", "Blocked"),
+    (re.compile(r"^🔒\s*"), "callout-good", "Guarantee"),
+    (re.compile(r"^📝\s*"), "callout-info", "Note"),
+    (re.compile(r"^🚨\s*"), "callout-bad", "Critical"),
+    (re.compile(r"^🛠\s*"), "callout-info", "Note"),
+    (re.compile(r"^🆕\s*"), "callout-info", "New"),
+    (re.compile(r"^⚙️⚡?\s*"), "callout-info", "Config"),
+    (re.compile(r"^🔁\s*"), "callout-info", "Loop"),
+    (re.compile(r"^🌱\s*"), "callout-info", "Note"),
+    (re.compile(r"^📚\s*"), "callout-info", "Reference"),
+    (re.compile(r"^🧬\s*"), "callout-info", "Meta"),
+]
+
+
+def _slugify(text: str) -> str:
+    text = re.sub(r"[`*_~$]", "", text)
+    text = re.sub(r"\s+", "-", text.strip().lower())
+    text = re.sub(r"[^a-z0-9一-鿿\-]", "", text)
+    return text.strip("-") or "section"
+
+
+def parse_blocks(lines: list[str]) -> list[dict]:
+    blocks: list[dict] = []
+    i = 0
+    n = len(lines)
+    while i < n:
+        line = lines[i]
+        stripped = line.rstrip()
+
+        # Blank line — skip.
+        if not stripped.strip():
+            i += 1
+            continue
+
+        # Code fence.
+        m = _RE_CODE_FENCE.match(stripped)
+        if m:
+            lang = (m.group("lang") or "").strip()
+            flags_raw = (m.group("flags") or "").strip()
+            flags: set[str] = set()
+            if flags_raw:
+                for tok in flags_raw.split(","):
+                    tok = tok.strip()
+                    if tok:
+                        flags.add(tok)
+            body: list[str] = []
+            i += 1
+            while i < n and not _RE_CODE_FENCE.match(lines[i].rstrip()):
+                body.append(lines[i])
+                i += 1
+            if i < n:  # consume closing fence
+                i += 1
+            blocks.append({
+                "type": "code",
+                "lang": lang,
+                "content": "\n".join(body),
+                "flags": flags,
+            })
+            continue
+
+        # ATX heading.
+        m = _RE_HEADING.match(stripped)
+        if m:
+            level = len(m.group(1))
+            text = m.group(2).strip()
+            blocks.append({"type": "heading", "level": level, "text": text})
+            i += 1
+            continue
+
+        # Horizontal rule.
+        if _RE_HR.match(stripped):
+            blocks.append({"type": "hr"})
+            i += 1
+            continue
+
+        # Table: current line has at least one `|`, AND next line is a divider.
+        if "|" in stripped and i + 1 < n and _RE_TABLE_DIVIDER.match(lines[i + 1].rstrip()):
+            header = stripped
+            divider = lines[i + 1].rstrip()
+            rows: list[str] = []
+            j = i + 2
+            while j < n and "|" in lines[j] and lines[j].strip():
+                rows.append(lines[j].rstrip())
+                j += 1
+            blocks.append({"type": "table", "header": header, "divider": divider, "rows": rows})
+            i = j
+            continue
+
+        # HTML block — passthrough until the closing tag is reached. Per
+        # CommonMark §4.6 (loosely): treat the block as ending on the first
+        # line that *starts* with the matching </tag>. We track depth only
+        # for opening/closing tags that appear at line start (after optional
+        # whitespace), so inline mentions like `<details>` inside backticks
+        # in a paragraph don't perturb the counter.
+        if _RE_HTML_BLOCK_OPEN.match(line):
+            tag_m = _RE_HTML_BLOCK_OPEN.match(line)
+            assert tag_m is not None
+            tag = tag_m.group(1).lower()
+            line_open_re = re.compile(rf"^\s*<{re.escape(tag)}(?=[\s/>])", re.IGNORECASE)
+            line_close_re = re.compile(rf"^\s*</{re.escape(tag)}(?=[\s>])", re.IGNORECASE)
+            body = [line]
+            depth = 1
+            # Self-closing tag on the same line: <tag .../>
+            if re.search(rf"<{re.escape(tag)}\b[^>]*/\s*>", line, re.IGNORECASE):
+                depth = 0
+            # Same-line full close: <tag>...</tag> or </tag> on same line as open
+            elif re.search(rf"</{re.escape(tag)}(?=[\s>])", line, re.IGNORECASE):
+                depth = 0
+            i += 1
+            while i < n and depth > 0:
+                cur = lines[i]
+                body.append(cur)
+                # Count nested opens/closes at line start only.
+                if line_open_re.match(cur):
+                    depth += 1
+                if line_close_re.match(cur):
+                    depth -= 1
+                i += 1
+            content = "\n".join(body)
+            # Sanitize: strip dangerous tags / event handlers / unsafe URL schemes.
+            content = sanitize_html(content)
+            blocks.append({"type": "html", "content": content})
+            continue
+
+        # Blockquote.
+        if _RE_BLOCKQUOTE.match(stripped):
+            quote_lines: list[str] = []
+            while i < n:
+                m_bq = _RE_BLOCKQUOTE.match(lines[i].rstrip())
+                if not m_bq:
+                    break
+                quote_lines.append(m_bq.group(1))
+                i += 1
+            blocks.append({"type": "blockquote", "lines": quote_lines})
+            continue
+
+        # Ordered or unordered list. Indent-aware: a marker at the same
+        # indent as the first item starts a sibling; a marker at deeper
+        # indent (or any deeper-indented continuation line) becomes part
+        # of the current item's content and is re-parsed recursively in
+        # render_list (so nested lists / paragraphs / code blocks work).
+        if _RE_ORDERED.match(stripped) or _RE_UNORDERED.match(stripped):
+            first_m = _RE_ORDERED.match(lines[i]) or _RE_UNORDERED.match(lines[i])
+            base_indent = len(first_m.group(1))
+            ordered = bool(_RE_ORDERED.match(lines[i]))
+            items: list[str] = []
+            current: list[str] = []
+            while i < n:
+                cur_line = lines[i]
+                # Blank line — peek ahead to decide if list continues.
+                if cur_line.strip() == "":
+                    if i + 1 < n:
+                        nxt = lines[i + 1]
+                        m_o_next = _RE_ORDERED.match(nxt)
+                        m_u_next = _RE_UNORDERED.match(nxt)
+                        if (m_o_next or m_u_next) and len(
+                            (m_o_next or m_u_next).group(1)
+                        ) >= base_indent:
+                            # Continue list; consume blank.
+                            current.append("")
+                            i += 1
+                            continue
+                        if nxt.startswith(" " * (base_indent + 2)) or nxt.startswith("\t"):
+                            current.append("")
+                            i += 1
+                            continue
+                    break
+                m_o = _RE_ORDERED.match(cur_line)
+                m_u = _RE_UNORDERED.match(cur_line)
+                if m_o or m_u:
+                    marker_indent = len((m_o or m_u).group(1))
+                    if marker_indent == base_indent:
+                        # Sibling item at same level.
+                        if current:
+                            items.append("\n".join(current).rstrip())
+                            current = []
+                        current.append((m_o or m_u).group(3 if m_o else 2))
+                        i += 1
+                        continue
+                    if marker_indent > base_indent:
+                        # Nested item — append the line with reduced indent
+                        # (strip base_indent + 2 spaces so the nested parse
+                        # sees the marker at column 0 of its own context).
+                        strip = base_indent + 2
+                        current.append(
+                            cur_line[strip:] if cur_line[:strip].strip() == "" else cur_line.lstrip()
+                        )
+                        i += 1
+                        continue
+                    # Marker at shallower indent — list ends here.
+                    break
+                # Non-marker line: continuation of current item if indented
+                # past base_indent + 2 (or any indent for the "lazy continuation"
+                # convention).
+                if cur_line.startswith(" " * (base_indent + 2)) or cur_line.startswith("\t"):
+                    strip = base_indent + 2
+                    current.append(
+                        cur_line[strip:] if cur_line[:strip].strip() == "" else cur_line.lstrip()
+                    )
+                    i += 1
+                    continue
+                # Anything else at base_indent or shallower → list ends.
+                break
+            if current:
+                items.append("\n".join(current).rstrip())
+            blocks.append({"type": "list", "ordered": ordered, "items": items})
+            continue
+
+        # Paragraph: gather lines until blank or block boundary.
+        para: list[str] = [stripped]
+        i += 1
+        while i < n:
+            nxt = lines[i].rstrip()
+            if not nxt.strip():
+                break
+            if (_RE_HEADING.match(nxt) or _RE_CODE_FENCE.match(nxt) or _RE_HR.match(nxt) or
+                _RE_BLOCKQUOTE.match(nxt) or _RE_ORDERED.match(nxt) or
+                _RE_UNORDERED.match(nxt) or _RE_HTML_BLOCK_OPEN.match(lines[i]) or
+                ("|" in nxt and i + 1 < n and _RE_TABLE_DIVIDER.match(lines[i + 1].rstrip()))):
+                break
+            para.append(nxt)
+            i += 1
+        blocks.append({"type": "paragraph", "text": " ".join(para)})
+    return blocks
+
+
+# ---------------------------------------------------------------------------
+# Block rendering
+# ---------------------------------------------------------------------------
+
+
+def render_table(header: str, divider: str, rows: list[str]) -> str:
+    def split_row(s: str) -> list[str]:
+        s = s.strip()
+        if s.startswith("|"):
+            s = s[1:]
+        if s.endswith("|"):
+            s = s[:-1]
+        return [c.strip() for c in s.split("|")]
+
+    header_cells = split_row(header)
+    align = []
+    for cell in split_row(divider):
+        cell = cell.strip()
+        if cell.startswith(":") and cell.endswith(":"):
+            align.append("center")
+        elif cell.endswith(":"):
+            align.append("right")
+        else:
+            align.append("left")
+
+    def cell_attr(idx: int) -> str:
+        if idx < len(align) and align[idx] != "left":
+            return f' style="text-align:{align[idx]}"'
+        return ""
+
+    out: list[str] = ["<table>", "<thead><tr>"]
+    for idx, cell in enumerate(header_cells):
+        out.append(f"<th{cell_attr(idx)}>{render_inline(cell)}</th>")
+    out.append("</tr></thead>")
+    out.append("<tbody>")
+    for row in rows:
+        out.append("<tr>")
+        for idx, cell in enumerate(split_row(row)):
+            out.append(f"<td{cell_attr(idx)}>{render_inline(cell)}</td>")
+        out.append("</tr>")
+    out.append("</tbody></table>")
+    return "".join(out)
+
+
+def render_blockquote(quote_lines: list[str]) -> str:
+    """Detect callout style (emoji prefix) and render with callout class."""
+    if not quote_lines:
+        return "<blockquote></blockquote>"
+    first = quote_lines[0]
+    css_class = None
+    title = None
+    body_lines = list(quote_lines)
+    # Try matching callout prefix on the first non-empty line.
+    first_nonempty_idx = 0
+    while first_nonempty_idx < len(body_lines) and not body_lines[first_nonempty_idx].strip():
+        first_nonempty_idx += 1
+    if first_nonempty_idx < len(body_lines):
+        line0 = body_lines[first_nonempty_idx].lstrip()
+        for pattern, klass, default_title in _CALLOUT_PREFIX_MAP:
+            m = pattern.match(line0)
+            if m:
+                css_class = klass
+                title = default_title
+                # Strip emoji + try to extract a **Bold:** title prefix.
+                rest = pattern.sub("", line0, count=1)
+                m_title = re.match(r"\*\*([^*\n]+?)\*\*[:：\s\-—]+\s*", rest)
+                if m_title:
+                    title = m_title.group(1)
+                    rest = rest[m_title.end():]
+                body_lines[first_nonempty_idx] = rest
+                break
+
+    inner_md = "\n".join(body_lines).strip("\n")
+    inner_blocks = parse_blocks(inner_md.split("\n")) if inner_md else []
+    inner_html, _ = _render_blocks(inner_blocks, collect_toc=False)
+    if css_class:
+        title_html = f'<div class="callout-title">{html_lib.escape(title or "")}</div>' if title else ""
+        return f'<div class="callout {css_class}">{title_html}{inner_html}</div>'
+    return f"<blockquote>{inner_html}</blockquote>"
+
+
+def render_list(ordered: bool, items: list[str]) -> str:
+    tag = "ol" if ordered else "ul"
+    out = [f"<{tag}>"]
+    for item in items:
+        # Nested lists are possible — parse the item body as blocks.
+        item_blocks = parse_blocks(item.split("\n"))
+        # Common case: single paragraph — emit inline directly.
+        if len(item_blocks) == 1 and item_blocks[0]["type"] == "paragraph":
+            out.append(f"<li>{render_inline(item_blocks[0]['text'])}</li>")
+        else:
+            inner, _ = _render_blocks(item_blocks, collect_toc=False)
+            out.append(f"<li>{inner}</li>")
+    out.append(f"</{tag}>")
+    return "".join(out)
+
+
+def render_code(lang: str, content: str, flags: set | None = None) -> str:
+    escaped = html_lib.escape(content)
+    # Per-block override for the auto-collapse JS in the template:
+    #   ```python {collapsed}  → force fold
+    #   ```python {open}       → force expanded
+    # Anything else leaves the decision to the auto-threshold (default 30 lines).
+    attr = ""
+    if flags:
+        if "collapsed" in flags:
+            attr = ' data-collapse="collapsed"'
+        elif "open" in flags:
+            attr = ' data-collapse="open"'
+    if lang:
+        return f'<pre{attr}><code class="language-{html_lib.escape(lang, quote=True)}">{escaped}</code></pre>'
+    # Heuristic: if content looks like an ASCII art diagram (mostly box-drawing
+    # chars and pipes), tag the surrounding <pre> with class="diagram".
+    diagram_chars = set("│─┌┐└┘├┤┬┴┼▲▼◀▶━┃┏┓┗┛╭╮╰╯═║╔╗╚╝╠╣╦╩╬║▶▼─")
+    sample = content[:200]
+    if sample and sum(1 for c in sample if c in diagram_chars) >= 4:
+        return f'<pre{attr} class="diagram"><code>{escaped}</code></pre>'
+    return f"<pre{attr}><code>{escaped}</code></pre>"
+
+
+def _render_blocks(
+    blocks: list[dict],
+    collect_toc: bool,
+    used_ids: dict | None = None,
+) -> tuple[str, list[dict]]:
+    if used_ids is None:
+        used_ids = {}
+    out: list[str] = []
+    toc: list[dict] = []
+    for b in blocks:
+        t = b["type"]
+        if t == "heading":
+            level = b["level"]
+            text = b["text"]
+            inline = render_inline(text)
+            base_id = _slugify(text)
+            uid = base_id
+            n = used_ids.get(base_id, 0)
+            if n > 0:
+                uid = f"{base_id}-{n}"
+            used_ids[base_id] = n + 1
+            out.append(f'<h{level} id="{uid}">{inline}</h{level}>')
+            if collect_toc and 2 <= level <= 3:
+                toc.append({"level": level, "id": uid, "text": text})
+        elif t == "paragraph":
+            out.append(f"<p>{render_inline(b['text'])}</p>")
+        elif t == "hr":
+            out.append("<hr />")
+        elif t == "code":
+            out.append(render_code(b.get("lang", ""), b["content"], b.get("flags")))
+        elif t == "blockquote":
+            out.append(render_blockquote(b["lines"]))
+        elif t == "list":
+            out.append(render_list(b["ordered"], b["items"]))
+        elif t == "table":
+            out.append(render_table(b["header"], b["divider"], b["rows"]))
+        elif t == "html":
+            out.append(_render_html_block(b["content"]))
+        else:
+            out.append(f"<!-- unknown block: {html_lib.escape(t)} -->")
+    return "\n".join(out), toc
+
+
+def _render_html_block(content: str) -> str:
+    """For <details>...</details> blocks, parse the inner content as Markdown.
+
+    Matches GitHub-flavored convention: when a `<details>` block has a blank
+    line separating its `<summary>` from the body, the body is parsed as
+    markdown (lists, headings, code, callouts all work). For non-<details>
+    HTML blocks, content passes through verbatim.
+    """
+    lines = content.split("\n")
+    if not lines or not lines[0].lstrip().lower().startswith("<details"):
+        return content
+    # Locate opening boundary: after </summary> if present, else after the
+    # <details> line itself.
+    open_end = 1  # default: just past the <details> line
+    summary_close_re = re.compile(r"</summary\s*>", re.IGNORECASE)
+    for j in range(1, len(lines)):
+        if summary_close_re.search(lines[j]):
+            open_end = j + 1
+            break
+        # If we hit non-summary content first, no `<summary>` block; stop scanning.
+        if lines[j].strip() and not re.match(r"\s*<summary", lines[j], re.IGNORECASE):
+            break
+    # Locate closing boundary: the last </details> line.
+    close_start = len(lines) - 1
+    close_re = re.compile(r"^\s*</details\s*>", re.IGNORECASE)
+    while close_start > 0 and not close_re.match(lines[close_start]):
+        close_start -= 1
+    if open_end >= close_start:
+        # Degenerate or empty body; emit raw.
+        return content
+    open_part = "\n".join(lines[:open_end])
+    inner_md = "\n".join(lines[open_end:close_start]).strip("\n")
+    close_part = "\n".join(lines[close_start:])
+    if not inner_md.strip():
+        return content
+    inner_blocks = parse_blocks(inner_md.split("\n"))
+    inner_html, _ = _render_blocks(inner_blocks, collect_toc=False)
+    return f"{open_part}\n{inner_html}\n{close_part}"
+
+
+def render_toc(toc: list[dict]) -> str:
+    """Render a nested TOC. Groups consecutive H3s under their preceding H2.
+
+    Orphan H3 (H3 without a preceding H2 in this run) is promoted to a
+    top-level <li>. Output is well-formed `<ol><li>...<ul><li>...</li></ul>...</li></ol>`.
+    """
+    if not toc:
+        return ""
+    grouped: list[tuple[dict, list[dict]]] = []  # [(h2_entry, [h3_children])]
+    current_parent: dict | None = None
+    current_children: list[dict] = []
+    for entry in toc:
+        if entry["level"] == 2:
+            if current_parent is not None:
+                grouped.append((current_parent, current_children))
+            current_parent = entry
+            current_children = []
+        elif entry["level"] == 3:
+            if current_parent is None:
+                # Orphan H3 — promote to top level so the TOC stays well-formed.
+                grouped.append((entry, []))
+            else:
+                current_children.append(entry)
+    if current_parent is not None:
+        grouped.append((current_parent, current_children))
+
+    out: list[str] = ["<ol>"]
+    for parent, children in grouped:
+        pid = html_lib.escape(parent["id"], quote=True)
+        ptext = html_lib.escape(parent["text"])
+        out.append(f'<li><a href="#{pid}">{ptext}</a>')
+        if children:
+            out.append("<ul>")
+            for c in children:
+                cid = html_lib.escape(c["id"], quote=True)
+                ctext = html_lib.escape(c["text"])
+                out.append(f'<li><a href="#{cid}">{ctext}</a></li>')
+            out.append("</ul>")
+        out.append("</li>")
+    out.append("</ol>")
+    return "\n".join(out)
+
+
+def strip_frontmatter(md: str) -> str:
+    """Strip leading YAML frontmatter (--- ... ---) if present at start.
+
+    Also strips a leading UTF-8 BOM so frontmatter detection still fires
+    on files saved by editors that prepend it.
+    """
+    md = md.lstrip("")
+    lines = md.split("\n", 1)
+    if not lines or lines[0].strip() != "---":
+        return md
+    # find closing ---
+    rest = lines[1] if len(lines) > 1 else ""
+    parts = rest.split("\n---\n", 1)
+    if len(parts) == 2:
+        return parts[1].lstrip("\n")
+    # try also \n--- at the very end of a line
+    parts = rest.split("\n---", 1)
+    if len(parts) == 2 and (parts[1] == "" or parts[1].startswith("\n")):
+        return parts[1].lstrip("\n")
+    return md
+
+
+# ---------------------------------------------------------------------------
+# Template loading + main
+# ---------------------------------------------------------------------------
+
+CDN_BLOCK_FULL = """<!-- MathJax 3 -->
+<script>
+window.MathJax = {
+  tex: { inlineMath: [['$', '$'], ['\\\\(', '\\\\)']], displayMath: [['$$', '$$'], ['\\\\[', '\\\\]']], processEscapes: true },
+  options: { skipHtmlTags: ['script', 'noscript', 'style', 'textarea', 'pre', 'code'] }
+};
+</script>
+<script src="https://cdn.jsdelivr.net/npm/mathjax@3/es5/tex-mml-chtml.js" async></script>
+
+<!-- highlight.js -->
+<link rel="stylesheet" href="https://cdn.jsdelivr.net/gh/highlightjs/cdn-release@11.9.0/build/styles/atom-one-light.min.css">
+<script src="https://cdn.jsdelivr.net/gh/highlightjs/cdn-release@11.9.0/build/highlight.min.js"></script>
+<script>document.addEventListener('DOMContentLoaded', () => hljs.highlightAll());</script>
+"""
+
+CDN_BLOCK_OFFLINE = "<!-- offline mode: MathJax + highlight.js skipped; math and code blocks render as plain text -->"
+
+
+def load_template(name: str) -> str:
+    path = TEMPLATES_DIR / f"{name}.html"
+    if not path.exists():
+        raise SystemExit(
+            f"error: template '{name}' not found at {path}. "
+            f"Available: {[p.stem for p in TEMPLATES_DIR.glob('*.html')]}"
+        )
+    return path.read_text(encoding="utf-8")
+
+
+def sha256_of(text: str) -> str:
+    return hashlib.sha256(text.encode("utf-8")).hexdigest()
+
+
+def render_json_as_pre(json_path: Path) -> str:
+    """Render a sidecar JSON file as a collapsed <details> block of pretty-printed JSON."""
+    try:
+        obj = json.loads(json_path.read_text(encoding="utf-8"))
+    except json.JSONDecodeError as e:
+        return f'<div class="callout callout-bad"><div class="callout-title">JSON parse error</div><p>{html_lib.escape(str(e))}</p></div>'
+    pretty = json.dumps(obj, ensure_ascii=False, indent=2, sort_keys=False)
+    return (
+        f'<details><summary>Sidecar JSON: <code>{html_lib.escape(str(json_path))}</code></summary>'
+        f'<pre><code class="language-json">{html_lib.escape(pretty)}</code></pre>'
+        f'</details>'
+    )
+
+
+def substitute(template: str, vars: dict) -> str:
+    out = template
+    for k, v in vars.items():
+        out = out.replace("{{" + k + "}}", v)
+    return out
+
+
+def json_for_script(obj: object) -> str:
+    """Serialize JSON for direct embedding inside a classic <script> block.
+
+    json.dumps() alone is unsafe because a string value containing '</script>'
+    will break out of the host <script> tag. Escape <, >, & and line/paragraph
+    separators as \\uXXXX so the result is always safe inside a <script>.
+    """
+    return (
+        json.dumps(obj, ensure_ascii=False)
+        .replace("&", "\\u0026")
+        .replace("<", "\\u003c")
+        .replace(">", "\\u003e")
+        .replace("
", "\\u2028")
+        .replace("
", "\\u2029")
+    )
+
+
+def _repo_relative(input_path: Path) -> str:
+    """Return a display-friendly repo-relative path. Avoids leaking absolute
+    /Users/<name>/... in the generated HTML meta + footer.
+
+    Order: cwd-relative if input is under cwd; else git-root-relative if in
+    a git repo; else basename only (parent dirs stripped) so we never
+    surface the home directory.
+    """
+    try:
+        return str(input_path.relative_to(Path.cwd()))
+    except ValueError:
+        pass
+    try:
+        import subprocess
+        result = subprocess.run(
+            ["git", "rev-parse", "--show-toplevel"],
+            cwd=input_path.parent,
+            capture_output=True,
+            text=True,
+            check=False,
+        )
+        if result.returncode == 0:
+            git_root = Path(result.stdout.strip())
+            try:
+                return str(input_path.relative_to(git_root))
+            except ValueError:
+                pass
+    except Exception:
+        pass
+    return input_path.name
+
+
+def main(argv: list[str] | None = None) -> int:
+    ap = argparse.ArgumentParser(
+        description="Render an ARIS Markdown artifact to single-file HTML.",
+    )
+    ap.add_argument("input", help="Path to input .md (or .json — wrapped in a <pre>)")
+    ap.add_argument("--template", default="academic", choices=["academic", "dashboard"])
+    ap.add_argument("--out", help="Output HTML path (default: <input>.html)")
+    ap.add_argument("--title", help="Page title (default: first H1, or filename)")
+    ap.add_argument("--subtitle", default="", help="Optional italic subtitle line")
+    ap.add_argument("--eyebrow", default="", help="Optional uppercase eyebrow above H1")
+    ap.add_argument("--author", default="", help="Optional author byline (e.g., 'Name (姓名), Affiliation')")
+    ap.add_argument("--lang", default="zh-CN", help='<html lang="…"> attribute (default zh-CN)')
+    ap.add_argument("--state", help="Optional sidecar state JSON to append as <details>")
+    ap.add_argument("--json", dest="json_sidecar", help="Optional sidecar JSON to append (e.g., KILL_ARGUMENT.json)")
+    ap.add_argument("--offline", action="store_true", help="Skip MathJax / highlight.js CDN blocks")
+    ap.add_argument("--no-toc", action="store_true", help="Skip TOC sidebar (forces TOC_LABEL/TOC_HTML to empty)")
+    ap.add_argument(
+        "--papers",
+        help="Sidecar JSON file with paper registry for [[key]] popovers. "
+             "Schema: {key: {title, authors, date, inst, key, arxiv}, ...}",
+    )
+    ap.add_argument(
+        "--blog-mode",
+        action="store_true",
+        help="Enable blog/talk mode (adds .aris-blog body class, opt-in active-H2 highlighting)",
+    )
+    ap.add_argument(
+        "--collapse-code-min",
+        type=int,
+        default=30,
+        help="Auto-collapse code blocks with >= N lines (default 30). Override per-block with ```lang {open}/{collapsed}.",
+    )
+    args = ap.parse_args(argv)
+
+    input_path = Path(args.input).resolve()
+    if not input_path.exists():
+        print(f"error: input not found: {input_path}", file=sys.stderr)
+        return 2
+    display_source_path = _repo_relative(input_path)
+
+    raw = input_path.read_text(encoding="utf-8")
+    source_hash = sha256_of(raw)
+
+    # If the input is JSON, wrap it as a single code block.
+    is_json = input_path.suffix.lower() == ".json"
+    if is_json:
+        try:
+            obj = json.loads(raw)
+            pretty = json.dumps(obj, ensure_ascii=False, indent=2, sort_keys=False)
+        except json.JSONDecodeError:
+            pretty = raw
+        md_source = f"# {input_path.name}\n\n```json\n{pretty}\n```\n"
+    else:
+        md_source = strip_frontmatter(raw)
+
+    blocks = parse_blocks(md_source.split("\n"))
+    body_html, toc = _render_blocks(blocks, collect_toc=not args.no_toc)
+
+    # Title autodetection.
+    title = args.title
+    if not title:
+        # Look for first H1 block.
+        for b in blocks:
+            if b.get("type") == "heading" and b.get("level") == 1:
+                title = b["text"]
+                break
+        if not title:
+            title = input_path.stem.replace("_", " ").replace("-", " ").title()
+
+    # Append sidecar JSON if requested.
+    extra_html_blocks: list[str] = []
+    for label, path_str in (("state", args.state), ("json", args.json_sidecar)):
+        if path_str:
+            p = Path(path_str).resolve()
+            if p.exists():
+                extra_html_blocks.append(f'<h2 id="sidecar-{label}">Sidecar — <code>{html_lib.escape(p.name)}</code></h2>')
+                extra_html_blocks.append(render_json_as_pre(p))
+            else:
+                extra_html_blocks.append(
+                    f'<div class="callout callout-warn"><div class="callout-title">Sidecar missing</div>'
+                    f'<p><code>{html_lib.escape(path_str)}</code> not found.</p></div>'
+                )
+
+    body_html = body_html + ("\n" + "\n".join(extra_html_blocks) if extra_html_blocks else "")
+
+    template_str = load_template(args.template)
+
+    toc_html = render_toc(toc) if not args.no_toc else ""
+    toc_label = "Contents" if not args.no_toc else ""
+    if args.no_toc:
+        # Hide TOC entirely by emitting an empty <nav> (template has <nav>, we'd
+        # rather just leave structure intact; CSS won't visually break).
+        toc_html = ""
+
+    eyebrow_block = f'<div class="eyebrow">{html_lib.escape(args.eyebrow)}</div>' if args.eyebrow else ""
+    subtitle_block = f'<p class="subtitle">{html_lib.escape(args.subtitle)}</p>' if args.subtitle else ""
+    byline_block = (
+        f'<p class="byline">By <strong>{html_lib.escape(args.author)}</strong></p>'
+        if args.author else ""
+    )
+
+    generated_at = datetime.now(timezone.utc).strftime("%Y-%m-%d %H:%M UTC")
+
+    # Paper registry sidecar (for [[key]] popovers). Validate as JSON before
+    # embedding; on parse error, log a warning and emit "{}" so the template
+    # still works (popover JS just no-ops when key is missing).
+    papers_json = "{}"
+    if args.papers:
+        p = Path(args.papers).resolve()
+        if p.exists():
+            try:
+                obj = json.loads(p.read_text(encoding="utf-8"))
+                papers_json = json_for_script(obj)
+            except json.JSONDecodeError as e:
+                print(f"warning: --papers JSON parse error: {e}", file=sys.stderr)
+        else:
+            print(f"warning: --papers file not found: {p}", file=sys.stderr)
+
+    vars_ = {
+        "LANG": html_lib.escape(args.lang, quote=True),
+        "TITLE": html_lib.escape(title),
+        "SUBTITLE_BLOCK": subtitle_block,
+        "EYEBROW_BLOCK": eyebrow_block,
+        "BYLINE_BLOCK": byline_block,
+        "SOURCE_PATH": html_lib.escape(display_source_path),
+        "SOURCE_SHA256": source_hash,
+        "SOURCE_SHA256_SHORT": source_hash[:12],
+        "GENERATED_AT": generated_at,
+        "HEAD_CDN": CDN_BLOCK_OFFLINE if args.offline else CDN_BLOCK_FULL,
+        "TOC_HTML": toc_html,
+        "TOC_LABEL": toc_label,
+        "BODY_HTML": body_html,
+        "EXTRA_META": "",
+        "PAPER_REGISTRY_JSON": papers_json,
+        "COLLAPSE_CODE_MIN": str(args.collapse_code_min),
+        "BODY_CLASS": "aris-blog" if args.blog_mode else "",
+    }
+
+    rendered = substitute(template_str, vars_)
+
+    out_path = Path(args.out).resolve() if args.out else input_path.with_suffix(".html")
+    if out_path.is_dir():
+        print(
+            f"error: --out points to a directory: {out_path}. "
+            f"Specify a file path ending in .html.",
+            file=sys.stderr,
+        )
+        return 2
+    out_path.parent.mkdir(parents=True, exist_ok=True)
+    out_path.write_text(rendered, encoding="utf-8")
+    print(f"wrote {out_path} ({len(rendered):,} bytes, {len(toc)} TOC entries, source sha256 {source_hash[:12]}...)")
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())