docs(agent-docs): enforce canonical AGENTS symlinks

.github/scripts/agent-docs-l1.mjs

@@ -28,6 +28,7 @@ export const CONFIG = {
     'tests/consumer-typecheck/node_modules',
   ],
   intentionalDifferentPairs: ['packages/superdoc/AGENTS.md:packages/superdoc/CLAUDE.md'],
+  canonicalSymlinkTarget: 'AGENTS.md',
   knownCommands: [
     'pnpm test',
     'pnpm test:behavior',
@@ -285,9 +286,15 @@ function classifyPairs(files) {
       pairs.push({ dir, classification: 'intentional-different', detail: `allowlisted: ${a.lineCount}L vs ${c.lineCount}L` });
       continue;
     }
-    const linked = (a.isSymlink && a.symlinkTarget === c.absPath) || (c.isSymlink && c.symlinkTarget === a.absPath);
-    if (linked) {
-      pairs.push({ dir, classification: 'linked', detail: `canonical: ${a.isSymlink ? 'CLAUDE.md' : 'AGENTS.md'}` });
+    const agentsPointsToClaude = a.isSymlink && a.symlinkTarget === c.absPath;
+    const claudePointsToAgents = c.isSymlink && c.symlinkTarget === a.absPath;
+    if (agentsPointsToClaude || claudePointsToAgents) {
+      const canonical = agentsPointsToClaude ? 'CLAUDE.md' : 'AGENTS.md';
+      const classification = canonical === CONFIG.canonicalSymlinkTarget ? 'linked' : 'linked-inverted';
+      const detail = canonical === CONFIG.canonicalSymlinkTarget
+        ? `canonical: ${canonical}`
+        : `canonical: ${canonical}; expected ${CONFIG.canonicalSymlinkTarget}`;
+      pairs.push({ dir, classification, detail });
       continue;
     }
     // Either side might be a broken symlink we already flagged; if so we
@@ -335,6 +342,12 @@ export function computeFlags(file) {
   return reasons;
 }
 
+function pairFlaggedForReview(pair) {
+  if (pair.classification === 'linked-inverted') return true;
+  if (pair.classification === 'unexpected-duplicate') return true;
+  return false;
+}
+
 // L2/L3 gating: stricter than computeFlags. A single-broken-ref doc still
 // appears in the L1 report (via computeFlags) but only triggers paid L2/L3
 // review when there are 2+ broken refs or budget is significantly exceeded.
@@ -392,9 +405,14 @@ export function renderL1Markdown(scan) {
     findingsByFile.push({ file: f, reasons });
   }
   lines.push('\n## Deterministic findings\n');
-  if (findingsByFile.length === 0) {
+  const pairFindings = scan.pairs.filter(pairFlaggedForReview);
+  if (findingsByFile.length === 0 && pairFindings.length === 0) {
     lines.push('None.\n');
   } else {
+    for (const pair of pairFindings) {
+      lines.push(`### \`${pair.dir || '(root)'}\`\n`);
+      lines.push(`- ${pair.classification}: ${pair.detail}\n`);
+    }
     for (const { file, reasons } of findingsByFile) {
       lines.push(`### \`${file.relPath}\`\n`);
       lines.push(reasons.map((r) => `- ${r}`).join('\n'));

AGENTS.md

@@ -0,0 +1,85 @@
+# SuperDoc
+
+A document editing and rendering library for the web.
+
+## Architecture: Rendering
+
+SuperDoc uses its own rendering pipeline. ProseMirror stores document state; it is not the visual renderer.
+
+```
+.docx
+  → super-converter parses OOXML into the hidden PM doc
+  → pm-adapter reads PM state and resolved styles
+  → FlowBlock[]
+  → layout-engine paginates
+  → ResolvedLayout
+  → DomPainter paints DOM
+```
+
+- `PresentationEditor` wraps a hidden ProseMirror `Editor`. Its contenteditable DOM is never shown. PresentationEditor bridges editor events into layout/paint state; do not resolve OOXML semantics there.
+- **DomPainter** (`layout-engine/painters/dom/`) owns all visual rendering.
+- Style-resolved properties flow `pm-adapter` → DomPainter. Do not style document content with PM decorations.
+
+### Where To Put Your Change
+
+| Concern | Where | Rule |
+|---|---|---|
+| DOCX import/export | `super-editor/src/editors/v1/core/super-converter/` | Parse and preserve OOXML, style refs, inline properties. Do not bake resolved formatting into direct attrs. |
+| Style cascade | `layout-engine/style-engine/` | Single source of truth for defaults, styles, conditional formatting, inline overrides. |
+| Static document visuals | `pm-adapter/` data + `layout-engine/painters/dom/` rendering | Feed typed data into DomPainter. Do not style static content with PM decorations. |
+| Direction-aware properties | `layout-engine/painters/dom/` | DomPainter mirrors at paint time for `w:bidiVisual`. pm-adapter stores logical sides LTR-default. Pre-mirroring upstream is a double-swap. See `packages/layout-engine/pm-adapter/src/direction/README.md`. |
+| Editing behavior | `super-editor/src/editors/v1/extensions/` | Commands, keybindings, editor plugins. Do not duplicate cascade or render document visuals here. |
+| Final DOM rendering | `layout-engine/painters/dom/` | Render `ResolvedLayout`. Paint-time transforms (e.g. RTL mirror) live here. |
+| New doc-api operation | `packages/document-api/src/contract/operation-definitions.ts` | Contract-first; touches 4 files. See `packages/document-api/README.md`. |
+
+For specialized boundaries (interaction mapping, geometry/pagination, ephemeral overlays, presentation state bridge, consumer SDK surface), see `packages/layout-engine/AGENTS.md` and the relevant package AGENTS.md.
+
+### Boundary check
+
+Before adding a visual or direction-aware path, run:
+
+```bash
+# Painter must not import upstream packages.
+rg "@superdoc/(pm-adapter|style-engine|layout-bridge|layout-resolved)" packages/layout-engine/painters/dom/src
+```
+
+More checks in `packages/layout-engine/AGENTS.md`.
+
+## Style Resolution Boundary
+
+The importer stores raw OOXML. The style-engine resolves at render time.
+
+- Converter (`super-converter/`) parses and stores only what is explicitly in the XML.
+- Style-engine (`layout-engine/style-engine/`) owns cascade logic.
+
+**Why**: resolving during import bakes inline properties into nodes; export then writes direct formatting instead of style references and loses document intent.
+
+## Document API Contract
+
+`packages/document-api/` uses a contract-first pattern.
+
+- **`operation-definitions.ts`** is the canonical object. All downstream maps project from it.
+- **`operation-registry.ts`** is the type-level registry (`input`, `options`, `output`).
+- **`invoke.ts`** is the dispatch table, validated against the registry at compile time.
+
+Adding an operation touches 4 files: `operation-definitions.ts`, `operation-registry.ts`, `invoke.ts`, and the implementation. Run `pnpm run generate:all` after. See `packages/document-api/README.md`.
+
+Do not hand-edit `COMMAND_CATALOG`, `OPERATION_MEMBER_PATH_MAP`, `OPERATION_REFERENCE_DOC_PATH_MAP`, or `REFERENCE_OPERATION_GROUPS`. They are derived from `OPERATION_DEFINITIONS`.
+
+## Commands
+
+- `pnpm build` - build all packages
+- `pnpm test` - unit tests
+- `pnpm dev` - dev server from `examples/`
+- `pnpm run generate:all` - regenerate schemas, SDK clients, tool catalogs, reference docs
+
+## Testing
+
+| What to verify | Command | Speed |
+|---|---|---|
+| Logic works? | `pnpm test` | seconds |
+| Editing works? | `pnpm test:behavior` | minutes |
+| Layout regressed? | `pnpm test:layout` | ~10 min |
+| Pixel diff? | `pnpm test:visual` | ~5 min |
+
+Per-package detail: `tests/behavior/AGENTS.md`, `tests/visual/AGENTS.md`. Eval suite: `evals/AGENTS.md`.

CLAUDE.md

@@ -0,0 +1 @@
+AGENTS.md

agent-docs-policy.md

@@ -13,6 +13,9 @@ encyclopedias of the codebase.
   Loaded only when an agent reads files in that package.
 - **`.claude/rules/<topic>.md`** with `paths:` frontmatter: rules that only
   apply to matching files (e.g. JSDoc rules for `**/*.js`).
+- **Symlink pairs**: when both `AGENTS.md` and `CLAUDE.md` exist with the same
+  content, make `AGENTS.md` canonical and symlink `CLAUDE.md` to it. Allowlist
+  intentional audience-specific pairs that must differ.
 - **Hooks or scripts**: anything that must be enforced rather than advised.
   Doc-level "always do X" rules that have a deterministic check belong in CI,
   not in agent docs.