emrg: fix Monaco AMD define hijacking UMD marked — markdown degraded to plain text (rant 2026-08-15T10:48:58)

[argszero]

Critical regression fix: after #801 (Monaco editor), chat and rant-detail markdown degraded to plain text.

Root cause (100% code-confirmed):

• monaco-editor 0.56.0's vendor/monaco/vs/loader.js defines GLOBAL AMD define/require on window (loader.js L1347 u.global.define = p).
• marked.min.js is UMD: it detects define.amd and takes the AMD define branch → window.marked is never set.
• index.html loaded monaco loader.js (L390) BEFORE marked/dompurify/highlight (L403-405) → window.marked empty → renderMarkdown/streamProject all degraded to escapeHtml plain text (chat + rant details).

Fixes:

1. index.html: marked/dompurify/highlight now load BEFORE monaco loader.js — no AMD present at their load time → UMD branch → window.marked set; Monaco's AMD define afterwards does not affect already-loaded globals.
2. markdown.js: defensive AMD-module-table recovery (recoverAmdModule("marked") reads monaco require.s/define.amd.modules) before degrading; upgraded the missing-vendor log to a console.error naming the AMD-hijack cause for future diagnosis.
3. Tests: build-config.test.js order guard (marked/dompurify/highlight must precede loader.js — fails if order regresses) + renderer.smoke AMD-recovery test (window.marked missing + AMD table present → renders via recovered marked; both missing → escapeHtml fallback).
4. Agent.md GUI count synced 253→255 (renderer smoke 127, build-config 7).

Verification: GUI npm test 255/255; pytest 824 passed/1 skipped; doc-count guard 3/3; import + CLI checks green. Rant details markdown (and its pending 【】-header enhancement) now render again once this lands.

[argszero]

✅ LGTM — cycle

[argszero]

✅ LGTM — cycle

Reviewed the full diff: root cause correctly identified (monaco loader.js defines global AMD define/require at L1347 → UMD marked takes the define branch → window.marked never set → plain-text degradation); index.html order fix (UMD vendors before monaco loader) is the right structural fix; recoverAmdModule() fallback reading require.s module table is a sound defensive layer with positive/negative test coverage; console.error diagnostic names the cause; order-guard test locks the fix against regression. CI test + test-windows both SUCCESS (run 31860446814). Agent.md counts synced (253→255).

[pm25coder]

I tested this PR end-to-end on Windows and the root-cause fix checks out.

[argszero]

✅ LGTM — cycle

Reviewed the full diff: script-order fix is the correct root-cause resolution (UMD marked sees AMD define from monaco loader → never sets window.marked; loading marked/dompurify/highlight first makes them take the UMD branch, and monaco's AMD define afterwards doesn't disturb already-set globals). The AMD module-table recovery in markdown.js is a sound defensive layer (require.s / define.amd.modules), and the console.error names the exact cause for future diagnosis. Order-guard test (build-config) + AMD-recovery test (renderer.smoke) both cover the regression; Agent.md counts synced (253→255). CI test + test-windows SUCCESS (run 31860446814).