@litfamily/litopencode 1.0.7 → 1.0.10

package/skills/lit-pptx/scripts/validate_pptx.py

@@ -0,0 +1,206 @@
+#!/usr/bin/env python3
+"""
+validate_pptx.py — lightweight anti-slop validator for PPTX outputs.
+
+Checks deck text hygiene, title presence, placeholder/default copy, forbidden
+terms, and coarse visual consistency signals. Prints JSON.
+"""
+
+from __future__ import annotations
+
+import json
+import re
+import sys
+from collections import Counter
+from pathlib import Path
+from typing import Any
+
+from pptx import Presentation
+from pptx.dml.color import RGBColor
+from pptx.enum.shapes import MSO_SHAPE_TYPE, PP_PLACEHOLDER
+
+
+SKILL_DIR = Path(__file__).resolve().parent.parent  # repo root (validate_pptx.py lives in scripts/)
+FORBIDDEN_PATH = SKILL_DIR / "FORBIDDEN_TERMS.json"
+DEFAULT_BAD_FONTS = {"Calibri", "Aptos"}
+DEFAULT_PLACEHOLDER_TEXT = {
+    "click to add title",
+    "click to add subtitle",
+    "click to add text",
+    "title",
+    "subtitle",
+}
+MAX_DISTINCT_FILLS = 8
+WEIGHTS = {
+    "missing_titles": 20,
+    "forbidden_terms": 20,
+    "default_placeholder_text": 20,
+    "default_fonts": 15,
+    "rainbow_fills": 10,
+}
+
+
+def _load_terms() -> tuple[list[str], list[str], bool, bool]:
+    if not FORBIDDEN_PATH.exists():
+        return [], [], True, False
+    data = json.loads(FORBIDDEN_PATH.read_text())
+    return (
+        data.get("terms", []),
+        data.get("soft_terms", []),
+        bool(data.get("case_insensitive", True)),
+        bool(data.get("whole_word", False)),
+    )
+
+
+def _build_regex(terms: list[str], case_insensitive: bool, whole_word: bool) -> re.Pattern[str] | None:
+    if not terms:
+        return None
+    escaped = [re.escape(term) for term in terms]
+    pattern = "|".join(escaped)
+    if whole_word:
+        pattern = rf"(?:(?<=\W)|^)(?:{pattern})(?:(?=\W)|$)"
+    flags = re.IGNORECASE if case_insensitive else 0
+    return re.compile(pattern, flags)
+
+
+def _shape_text(shape) -> str:
+    if not getattr(shape, "has_text_frame", False):
+        return ""
+    parts: list[str] = []
+    for paragraph in shape.text_frame.paragraphs:
+        chunk = "".join(run.text for run in paragraph.runs).strip()
+        if chunk:
+            parts.append(chunk)
+    return "\n".join(parts).strip()
+
+
+def _shape_fill(shape) -> str | None:
+    fill = getattr(shape, "fill", None)
+    if fill is None or fill.type is None:
+        return None
+    try:
+        fore = getattr(fill, "fore_color", None)
+    except TypeError:
+        return None
+    rgb = getattr(fore, "rgb", None)
+    if isinstance(rgb, RGBColor):
+        return str(rgb)
+    if isinstance(rgb, str):
+        return rgb.upper()
+    return None
+
+
+def _iter_runs(slide):
+    for shape in slide.shapes:
+        if not getattr(shape, "has_text_frame", False):
+            continue
+        for paragraph in shape.text_frame.paragraphs:
+            for run in paragraph.runs:
+                yield shape, run
+
+
+def _slide_has_title(slide, slide_height: int | None) -> bool:
+    title_shape = getattr(slide.shapes, "title", None)
+    if title_shape is not None and _shape_text(title_shape):
+        return True
+    for shape in slide.shapes:
+        if not getattr(shape, "is_placeholder", False):
+            continue
+        placeholder = getattr(shape, "placeholder_format", None)
+        if placeholder and placeholder.type == PP_PLACEHOLDER.TITLE and _shape_text(shape):
+            return True
+    text_shapes = []
+    for shape in slide.shapes:
+        if not getattr(shape, "has_text_frame", False):
+            continue
+        text = _shape_text(shape)
+        if not text:
+            continue
+        text_shapes.append(shape)
+        top = getattr(shape, "top", None)
+        if slide_height is not None and top is not None and top <= slide_height * 0.22:
+            return True
+    if len(text_shapes) <= 2 and any(_shape_text(shape) for shape in text_shapes):
+        return True
+    return False
+
+
+def lint(path: str | Path) -> dict[str, Any]:
+    prs = Presentation(str(path))
+    hard_terms, soft_terms, case_insensitive, whole_word = _load_terms()
+    hard_re = _build_regex(hard_terms, case_insensitive, whole_word)
+    soft_re = _build_regex(soft_terms, case_insensitive, whole_word)
+
+    missing_titles: list[int] = []
+    placeholder_hits: list[str] = []
+    hard_hits: list[str] = []
+    soft_hits: list[str] = []
+    default_font_hits: list[str] = []
+    fills: Counter[str] = Counter()
+
+    for slide_idx, slide in enumerate(prs.slides, start=1):
+        if not _slide_has_title(slide, prs.slide_height):
+            missing_titles.append(slide_idx)
+
+        for shape_idx, shape in enumerate(slide.shapes, start=1):
+            fill = _shape_fill(shape)
+            if fill:
+                fills[fill] += 1
+
+            text = _shape_text(shape)
+            if not text:
+                continue
+
+            lowered = text.strip().lower()
+            if lowered in DEFAULT_PLACEHOLDER_TEXT:
+                placeholder_hits.append(f"slide {slide_idx} shape {shape_idx}: {text[:80]}")
+
+            if hard_re and hard_re.search(text):
+                hard_hits.append(f"slide {slide_idx} shape {shape_idx}: {text[:120]}")
+            elif soft_re and soft_re.search(text):
+                soft_hits.append(f"slide {slide_idx} shape {shape_idx}: {text[:120]}")
+
+        for shape, run in _iter_runs(slide):
+            font_name = (run.font.name or "").strip()
+            if font_name in DEFAULT_BAD_FONTS:
+                default_font_hits.append(f"slide {slide_idx}: {font_name}")
+
+    last_slide_text = "\n".join(_shape_text(shape) for shape in prs.slides[-1].shapes if getattr(shape, "has_text_frame", False)).lower() if prs.slides else ""
+    has_source_note = "source:" in last_slide_text or "출처" in last_slide_text
+
+    failures = {
+        "missing_titles": bool(missing_titles),
+        "forbidden_terms": bool(hard_hits),
+        "default_placeholder_text": bool(placeholder_hits),
+        "default_fonts": bool(default_font_hits),
+        "rainbow_fills": len(fills) > MAX_DISTINCT_FILLS,
+    }
+    slop_score = sum(WEIGHTS[key] for key, failed in failures.items() if failed)
+
+    return {
+        "file": str(path),
+        "slide_count": len(prs.slides),
+        "pass": slop_score <= 10 and not hard_hits and not placeholder_hits,
+        "slop_score": slop_score,
+        "checks": {
+            "missing_titles": {"pass": not missing_titles, "slides": missing_titles},
+            "forbidden_terms": {"pass": not hard_hits, "hard_hits": hard_hits[:20], "soft_warnings": soft_hits[:20]},
+            "default_placeholder_text": {"pass": not placeholder_hits, "hits": placeholder_hits[:20]},
+            "default_fonts": {"pass": not default_font_hits, "hits": default_font_hits[:20]},
+            "rainbow_fills": {"pass": len(fills) <= MAX_DISTINCT_FILLS, "distinct_fills": len(fills), "sample": list(fills.keys())[:12]},
+            "missing_source_note": {"pass": has_source_note},
+        },
+    }
+
+
+def main(argv: list[str]) -> int:
+    if len(argv) != 2:
+        print(f"Usage: {Path(argv[0]).name} <file.pptx>", file=sys.stderr)
+        return 2
+    report = lint(argv[1])
+    print(json.dumps(report, ensure_ascii=False, indent=2))
+    return 0 if report["pass"] else 1
+
+
+if __name__ == "__main__":
+    raise SystemExit(main(sys.argv))

package/skills/lit-pptx/specs/markdown-slide-spec-v1.md

@@ -0,0 +1,251 @@
+# Markdown Slide Specification v1
+
+> **Status:** Superseded by `markdown-slide-spec-v2.md`, and still valid. This document
+> defines the v1 contract; a deck written entirely to it compiles to a `slide-ast-v1`
+> document exactly as before. v2 is additive — read it for placement (`region`,
+> `columns`, `box`, `shape`), variants, and the open layout set. The one non-goal v2
+> reverses is §11's ban on free-form absolute positioning.
+> **Date:** 2026-04-19
+
+The executive/research profile below is a backward-compatible authoring profile. It composes existing v1 blocks and does not add coordinates, CSS, directives, or AST types.
+
+## Purpose
+
+Define a constrained Markdown + directive dialect that serves as the single authoring source for PPTX slide decks. The dialect must be:
+
+1. **LLM-friendly** — predictable structure, easy for agents to generate and edit.
+2. **Precision-first** — compiles into a versioned intermediate AST that preserves layout intent.
+3. **Template-aware** — resolves against enrolled template contracts, not hardcoded behavior.
+
+## 1. Deck Structure
+
+A deck source file is a single Markdown file with:
+
+1. **YAML frontmatter** (required) — deck-level metadata enclosed in `---`.
+2. **Slide sections** — separated by `---` horizontal rules. Each section contains a layout directive and slide content.
+
+```
+---
+template: AZURE-PRO
+title: Deck Title
+date: 2026.04.19
+department: Department
+presenter: Name
+---
+
+---
+layout: cover
+
+# Cover Title
+---
+
+---
+layout: content
+
+## Section Title
+
+- Bullet 1
+- Bullet 2
+---
+```
+
+## 2. YAML Frontmatter
+
+Required fields:
+
+| Field | Type | Required | Description |
+|-------|------|----------|-------------|
+| `template` | string | yes | Enrolled template name (e.g. `AZURE-PRO`) |
+| `title` | string | yes | Deck title |
+| `date` | string | no | Presentation date |
+| `department` | string | no | Department name |
+| `presenter` | string | no | Presenter name |
+
+## 3. Slide Separators
+
+Slides are delimited by `---` on its own line (standard Markdown horizontal rule). The content between two `---` markers is a single slide section.
+
+Each slide section starts with a `layout` directive on the first non-empty line, followed by the slide content.
+
+## 4. Allowed Markdown Subset
+
+### 4.1 Permitted constructs
+
+| Construct | Syntax | Notes |
+|-----------|--------|-------|
+| Headings | `#`, `##`, `###` | Level 1 for cover/closing titles, Level 2 for slide titles, Level 3 for sub-headings |
+| Bullet lists | `- item` | Level 1 bullets |
+| Bold text | `**text**` | Within list items for section headers |
+| Numbered items | `(1) text` | Manual numbering with parenthesized Arabic numerals |
+| Tables | `\| ... \|` | Standard Markdown pipe tables |
+| Images | `![alt](path)` | With relative or absolute paths |
+| Table links | `[label](https://...)` | Rendered as a clickable external PPTX hyperlink in table cells |
+| Fenced blocks | `::: type ... :::` | Directive blocks (see §5) |
+
+### 4.2 Forbidden constructs
+
+The following are **rejected at parse time** with explicit errors:
+
+| Forbidden | Reason |
+|-----------|--------|
+| Raw HTML (`<div>`, `<span>`, etc.) | Bypasses template layout contracts |
+| `<style>` blocks | CSS escapes break PPTX fidelity |
+| Inline `style=` attributes | Arbitrary positioning breaks layout |
+| Nested fenced blocks (`:::` inside `:::`) | Ambiguous directive nesting |
+| Arbitrary CSS classes | Layout must come from template contracts |
+
+## 5. Directive Blocks
+
+Directives use fenced blocks with `:::` delimiters. Each directive type has a specific contract.
+
+### 5.1 v1 Directive Set
+
+| Directive | Purpose | Allowed in layouts |
+|-----------|---------|--------------------|
+| `main-box` | Bordered text box at bottom of slide | `main` |
+| `summary-group` | Grouped summary content (top/bottom) | `summary` |
+| `kpi-table` | Key Performance Indicator table | `main`, `content` |
+| `image` | Positioned image with caption | `content`, `main` |
+| `notes` | Speaker/footnotes | `content`, `main` |
+
+### 5.2 Directive Syntax
+
+```
+::: main-box
+Text content for the bordered box.
+:::
+```
+
+```
+::: kpi-table
+| Metric | Target | Actual |
+|--------|--------|--------|
+| Revenue | 5000억 | 5450억 |
+:::
+```
+
+```
+::: notes
+Source: Internal data, Q1 2026.
+:::
+```
+
+### 5.3 Nesting Rules
+
+- Directive blocks **MUST NOT** be nested inside other directive blocks.
+- Each `:::` open must have a matching `:::` close before another `:::` open.
+- Violation produces a parse error with the offending line number.
+
+## 6. Layout Types
+
+### 6.1 v1 Approved Layouts
+
+| Layout | Purpose | Required blocks |
+|--------|---------|-----------------|
+| `cover` | Opening slide | title (level 1) |
+| `content` | Standard title + body | title (level 2), body |
+| `main` | Content with optional bordered box | title (level 2), body, optional `main-box` |
+| `summary` | Two-group hierarchical summary | title (level 2), summary-groups via `###` headings |
+| `closing` | Thank-you slide | title (level 1) |
+
+### 6.2 Layout-specific content rules
+
+- `cover`: Exactly one `#` heading. No body lists.
+- `content`: One `##` heading. Body with bullet lists or tables.
+- `main`: One `##` heading. Body with bullet lists. Optional `::: main-box`.
+- `summary`: One `##` heading. Content organized under `###` sub-headings. The first `###` group maps to the top position, the second maps to the bottom position.
+- `closing`: Exactly one `#` heading. No body content.
+
+## 7. Bullet Hierarchy
+
+The template uses a two-level hierarchy:
+
+**Level 1 — Section headers:**
+```
+- **Section Title**
+```
+Rendered as bold bullet with `•` character.
+
+**Level 2 — Numbered sub-items:**
+```
+  (1) Detail item 1
+  (2) Detail item 2
+```
+Rendered as indented numbered items with `(1)`, `(2)` format.
+
+Indentation for level 2 items is 2 spaces.
+
+## 8. Failure Behavior
+
+### 8.1 Parse errors
+
+When the parser encounters a forbidden construct:
+
+1. **Stop parsing immediately** — do not silently skip or degrade.
+2. **Report the file path**, slide index, and approximate **line number**.
+3. **Name the specific violation** (e.g., "arbitrary HTML `<div>` not allowed", "nested directives at line 42").
+
+### 8.2 Validation errors
+
+When the compiled AST violates the schema:
+
+1. **Reject the entire deck** — do not produce partial output.
+2. **Report which slide and which block** failed validation.
+3. **Name the schema constraint** that was violated.
+
+## 9. AST Output Contract
+
+The compiler emits a JSON object conforming to `slide-ast-v1.schema.json`.
+
+Required top-level fields:
+
+| Field | Type | Description |
+|-------|------|-------------|
+| `astVersion` | string | Must be `"slide-ast-v1"` |
+| `version` | string | Schema version (semver) |
+| `deck` | object | Template name, title, metadata |
+| `slides` | array | Ordered array of slide objects |
+
+Each slide object must have:
+
+| Field | Type | Description |
+|-------|------|-------------|
+| `index` | number | 0-based slide position |
+| `layout` | string | One of the approved layout types |
+| `blocks` | array | Content blocks in order |
+
+Each block must have:
+
+| Field | Type | Description |
+|-------|------|-------------|
+| `type` | string | Block type: `title`, `body`, `main-box`, `summary-group`, `kpi-table`, `image`, `notes` |
+| `content` or `items` | varies | Block payload |
+
+## 10. Relationship to Enrolled Templates
+
+The AST is **template-independent**. It describes intent (layout, content structure), not pixel positions. Template-specific rendering (fonts, colors, positions, decorations) is resolved by the template registry and layout resolver at render time.
+
+This means:
+- The same AST can theoretically render through different enrolled templates.
+- Template enrollment contracts declare which layouts and blocks they support.
+- A deck targeting an unsupported layout/block for a given template produces a clear error, not a fallback.
+
+### 10.1 Evidence-oriented compositions
+
+Enrolled templates may advertise these semantic compositions of existing blocks:
+
+- **figure plus interpretation:** `image` + `figure-caption` + body interpretation;
+- **table plus decision takeaway:** table/`kpi-table` + `table-caption` + body takeaway;
+- **source capture:** `image` + `figure-caption` containing locator/source + body limitation/relevance;
+- **two-record appendix evidence:** title/body or `summary-group`, with no more than two records and DOI/canonical links when available.
+
+For a Markdown image used as evidence, alt text is the visible figure caption and should contain a figure number plus `Source:`/`출처:`. A blockquote immediately after a table is its visible table caption/source. Appendix DOI/canonical/raw URL fields use `[label](https://...)` in table cells so the PPTX contains an external hyperlink relationship. These requirements reuse existing parser and renderer behavior.
+
+## 11. Explicit Non-Goals for v1
+
+- Full Marp compatibility.
+- Arbitrary CSS as part of authoring input.
+- Free-form absolute positioning in Markdown.
+- Silent fallback or degradation for unsupported constructs.
+- Raw HTML passthrough.
+- JavaScript or interactive content.

package/skills/lit-pptx/specs/markdown-slide-spec-v2.md

@@ -0,0 +1,223 @@
+# Markdown Slide Specification v2
+
+> **Status:** Current. Additive over v1 — every v1 deck is a valid v2 deck and still
+> compiles to a `slide-ast-v1` document.
+> **Supersedes:** `markdown-slide-spec-v1.md`, which remains accurate for the subset
+> it describes and is still the contract v1 decks are held to.
+> **Date:** 2026-08-02
+
+## What changed and why
+
+v1 froze one non-goal that turned out to be the binding constraint in real work:
+*free-form absolute positioning in Markdown* (v1 §11). Combined with a closed set of
+six layouts and a fixed block-to-region map, a deck could only look like the shapes
+its template had anticipated. Building a cover the template did not already offer, or
+putting two tables on one slide, was not difficult — it was unexpressible.
+
+v2 opens placement in three tiers, so an author reaches for the least freedom that
+says the thing rather than dropping straight to coordinates.
+
+Nothing was removed. The v1 constructs, layouts, directives, separator rules and
+failure behaviour are unchanged, and a deck that uses none of §2–§6 below compiles to
+a byte-identical `slide-ast-v1` document.
+
+## 1. Version selection
+
+The compiler chooses the AST version from the deck's content, not from a flag.
+
+| Deck uses | `astVersion` |
+|---|---|
+| v1 constructs only | `slide-ast-v1` |
+| any of: `region`, `columns`, `box`, `shape`, `chart`, a slide-local key, a layout outside the v1 six | `slide-ast-v2` |
+
+An editable numeric chart uses a pipe table inside `::: chart type=bar` or
+`::: chart type=line`. The first column supplies category labels; remaining
+columns are numeric series. At least two category rows are required. Example:
+
+```
+::: chart type=bar
+| 분기 | 매출 | 이익 |
+| --- | ---: | ---: |
+| 1Q | 120 | 18 |
+| 2Q | 135 | 22 |
+:::
+```
+
+The PPTX route writes a native editable chart. Use `::: kpi-table` for a few
+headline metrics, with labels in the header row and values in the one data row.
+
+Both are accepted by the validator. This is what lets v1 fixtures and their assertions
+stand untouched.
+
+## 2. Slide-local keys
+
+Keys may follow the `layout:` line, ending at the first blank line:
+
+```
+---
+layout: cover
+variant: split-navy
+
+# Title
+---
+```
+
+**The key set is closed.** `variant` (§6) is the only slide-local key, and the run stops
+at the first line that is not one of them. A line the engine does not act on stays in
+the body.
+
+That is the whole guarantee, and it is deliberately not "anything shaped like a key".
+A body line reading `source: internal data` has exactly the shape of a key, and
+consuming it would delete it from the slide with no error. The same rule keeps a
+mistyped `varient:` visible in the body rather than turning it into an ignored key that
+silently does nothing — a typo you can see beats a variant that never applied.
+
+An earlier draft of this section claimed the blank line was the boundary and that prose
+was therefore safe. It was not. The Korean example it used, `출처: 내부 자료`, survived
+only because `출처` is not ASCII; the English equivalent was swallowed.
+
+## 3. Directive attributes
+
+A directive line may carry positional values and `key=value` pairs:
+
+```
+::: box x=6.9 y=2.6 w=5.3 h=1.4 role="body text" z=over
+```
+
+Numbers and booleans are coerced; quoted values keep their spaces. A v1 directive with
+no arguments parses exactly as before, with an empty attribute map.
+
+## 4. Nesting by fence length
+
+v1 forbade `:::` inside `:::` and every v1 deck depends on that. Rather than relax it,
+containers nest by fence length — the pandoc fenced-div convention. A block closes on a
+run of its own length, so `::::` may hold `:::` with no ambiguity.
+
+- Same-length nesting: **error** (unchanged from v1)
+- A longer fence inside a shorter one: **error**
+- Depth beyond two: **error**
+
+## 5. Placement
+
+### 5.1 Tier 1 — `region`
+
+```
+::: region name=right_figure
+| 항목 | 값 |
+|------|-----|
+:::
+```
+
+The body is parsed as ordinary Markdown; each resulting block is routed to the named
+region instead of the one its type would imply. The block list stays flat — `region`
+does not appear in the AST, it only sets `regionHint`.
+
+A name the layout does not declare is an error listing the regions it does.
+
+### 5.2 Tier 2 — `columns`
+
+```
+:::: columns 2fr 1fr gap=0.3
+::: col
+- 왼쪽
+:::
+::: col
+![도1. 개요](fig.png)
+:::
+::::
+```
+
+Tracks are ratios and the engine divides the layout's **body region** between them, so
+the source carries no coordinates and the same deck still lands correctly on a template
+with another canvas. `gap` is in inches, default `0.25`.
+
+- Track count must equal `::: col` count — a mismatch is an error, never silent padding.
+- A layout with no body region cannot host columns.
+- Columns replace the body region for that slide; the body region does not also render.
+
+### 5.3 Tier 3 — `box` and `shape`
+
+```
+::: box x=6.9 y=2.6 w=5.3 h=1.4 role=body_text z=over
+**결론** — 세 축으로 재편
+:::
+
+::: shape rect x=0 y=6.4 w=13.333 h=1.1 fill=#EEF4FF z=under
+```
+
+`box` requires `x y w h` in inches. `role` names a `font_role` from the template's
+`capabilities.yaml`, defaulting to `body_text`. Content is parsed as Markdown and flows
+down the box: every block but the last takes the height its content needs, the last
+takes what is left.
+
+`shape` is a **void directive** — one line, no closing fence. Its first positional value
+is the primitive: `rect · roundRect · pill · ellipse · ring · line · text · image`. Style
+attributes are `fill line lw radius rotate text color size align valign char_spacing
+bold asset dir src`, checked at compile time so a typo fails loudly. `line` is the one
+kind that needs no height; it takes its thickness from the stroke.
+
+`z` is `under` or `over`. Shapes default to `under`, boxes to `over`.
+
+### 5.4 What placement does not do
+
+**A template region is never reflowed to avoid a placed box.** The auto-nudge rules that
+narrow a body beside an image do not see placements at all, and coordinates arrive at the
+renderer exactly as written. Covering a region is therefore a defect the QA gate reports,
+not a nudge the engine applies. This is deliberate: an engine that quietly moved content
+out from under an author's box would make the author's intent unknowable.
+
+## 6. Layouts and variants
+
+The layout set is no longer closed. `spec-validator` checks that a layout name is an
+identifier (`^[a-z][a-z0-9-]*$`); whether it exists is a fact about the chosen template,
+and `layout-resolver` compares it against the template mapping and names the alternatives
+when it does not match.
+
+`free` needs no declaration — it means "this slide uses no template region", so every
+template supports it and the slide is built entirely from placements.
+
+A **variant** swaps a layout's decoration set and may override the regions the new
+furniture displaces:
+
+```
+---
+layout: cover
+variant: split-navy
+---
+```
+
+Both parts travel together, because a split cover needs a narrower title than a
+full-bleed one. An unknown variant is an error listing the ones the layout declares.
+
+## 7. Capability derivation
+
+When a template declares a layout but publishes no `supported_blocks` entry for it, the
+allowed blocks are derived from the regions that layout declares. Placement blocks and
+`notes` are allowed everywhere, since they do not depend on a region existing.
+
+## 8. Region occupancy
+
+A region holds one block. Two blocks claiming the same region is an error naming both
+ways out — `::: region name=` or `:::: columns`. In v1 the second block silently
+overwrote the first, which is why a slide could not carry two tables.
+
+Two exceptions are documented behaviour decks already rely on: images collapse into a
+grid, and captions are subordinate labels where the last one wins.
+
+## 9. Charts
+
+Charts are outside this dialect. A chart is produced by `/scientific-visualization` and
+placed as an image. `compile-deck.js --export-viz-context <path>` writes the template's
+palette, body font family, **bundled font file paths**, canvas and DPI so the chart is
+drawn in the deck's own visual system rather than a library default.
+
+## 10. Non-goals retained from v1
+
+- Full Marp compatibility
+- Arbitrary CSS as authoring input
+- Raw HTML passthrough
+- JavaScript or interactive content
+- Silent fallback or degradation for unsupported constructs
+
+The single v1 non-goal v2 reverses is free-form absolute positioning, and §5.4 states
+what was kept in exchange.