feat: stabilize TextMate syntax rendering

.codex/PIPELINE.md

@@ -58,6 +58,25 @@ Codex must obey:
 * VMLab moved to v0.3.5 and is blocked until the v0.3.0 Extensions Foundation is
   complete.
 
+## Current maintainer priority order
+
+For further ECLI work, Codex must sequence planning and implementation as:
+
+1. Stabilize all Extensions and the Extensions Layer.
+2. Finish TextMate rendering, TextMate performance, and multiline comment
+   handling.
+3. Implement F4 Diagnostics / Linter Panel.
+4. Implement F7 AI Code Assistant as an Extensions Layer feature.
+5. Implement the plugin layer for new themes, linters, and AI extensions.
+6. Implement F11 Terminal-console without duplicating PySH.
+7. Implement full mouse support.
+8. Implement F8 System Doctor / lab engineer diagnostic tools.
+9. Implement F9 Git improvements.
+
+This order does not override Stage gates, audit gates, the single-writer screen
+invariant, Extensions Layer security constraints, or release/publish safety
+rules.
+
 ## Recommended Codex execution mode
 
 For Stage 1 inventory and diagnostics:

AGENTS.md

@@ -691,3 +691,26 @@ All agents (Codex, Claude Code, Cursor, and any other automation) must obey:
   and all active platform packaging contracts must remain green.
 * VMLab is out of scope: it moved to v0.3.5 and is blocked until the v0.3.0
   Extensions Foundation is complete.
+
+## 14. Current maintainer priority order
+
+For further ECLI work, treat extension stabilization as the first priority:
+
+1. Stabilize all Extensions and the Extensions Layer before adjacent feature
+   expansion.
+2. Finish TextMate rendering, TextMate performance, and multiline comment
+   handling.
+3. Implement F4 Diagnostics / Linter Panel.
+4. Implement F7 AI Code Assistant as an Extensions Layer feature.
+5. Implement the plugin layer for new themes, linters, and AI extensions.
+6. Implement F11 Terminal-console without duplicating PySH. Command execution
+   must remain routed through explicit ECLI services, PySH, or CommandPlan
+   surfaces; do not migrate PySH source into ECLI.
+7. Implement full mouse support.
+8. Implement F8 System Doctor / lab engineer diagnostic tools.
+9. Implement F9 Git improvements.
+
+This priority order is a sequencing rule for planning and implementation. It
+does not waive Stage 1/Stage 2 gates, P0 audit requirements, the single-writer
+screen invariant, the Extensions Layer security contract, or release/publish
+safety rules.

CODEX.md

@@ -223,6 +223,30 @@ Codex must obey:
 * VMLab is out of scope: it moved to v0.3.5 and is blocked until the v0.3.0
   Extensions Foundation is complete.
 
+## Current maintainer priority order
+
+For further ECLI work, Codex must treat extension stabilization as the first
+priority:
+
+1. Stabilize all Extensions and the Extensions Layer before adjacent feature
+   expansion.
+2. Finish TextMate rendering, TextMate performance, and multiline comment
+   handling.
+3. Implement F4 Diagnostics / Linter Panel.
+4. Implement F7 AI Code Assistant as an Extensions Layer feature.
+5. Implement the plugin layer for new themes, linters, and AI extensions.
+6. Implement F11 Terminal-console without duplicating PySH. Command execution
+   must remain routed through explicit ECLI services, PySH, or CommandPlan
+   surfaces; do not migrate PySH source into ECLI.
+7. Implement full mouse support.
+8. Implement F8 System Doctor / lab engineer diagnostic tools.
+9. Implement F9 Git improvements.
+
+This priority order guides planning and implementation only. It does not waive
+Stage 1/Stage 2 gates, P0 audit requirements, the single-writer screen
+invariant, the Extensions Layer security contract, or release/publish safety
+rules.
+
 ## Expected Codex final response
 
 For non-trivial work, finish with:

README.md

@@ -673,6 +673,13 @@ Found a bug? Please help us by opening an issue on GitHub:
 
 * **Terminal**: Supports 256 colors and UTF-8
 
+* **TextMate syntax highlighting** (default): installed automatically via the
+  `python-textmate` dependency, which pulls `onigurumacffi` (the **Oniguruma**
+  regex engine). Binary wheels cover Linux, macOS, and Windows. On source-build
+  platforms (e.g. FreeBSD) install the Oniguruma development headers
+  (`devel/oniguruma`). If unavailable, ECLI falls back to the built-in legacy
+  highlighter automatically (`[extensions].syntax_engine = "legacy"`).
+
 ### Supported Platforms
 
 * Ubuntu 20.04 LTS and newer

audit-report.md

@@ -595,3 +595,54 @@ find .github/workflows -maxdepth 1 -type f -print | sort
 nl -ba .github/workflows/*.yml | sed -n '1,260p'
 rg -n "config|ConfigService|syntax_highlighting|History|undo|redo|runtime_import|pyinstaller|artifact|freebsd|coderabbit|CodeRabbit" tests .github docs/quality docs/release pyproject.toml
 ```
+
+## Issue #102 addendum — multiline TextMate protection
+
+Implementation:
+
+- ECLI keeps TextMate scopes as the primary token source for extension-backed
+  rendering.
+- `src/ecli/extensions/ecli_integration/syntax_service.py` now applies a
+  deterministic protected-range pass for known stateless multiline gaps:
+  Python strings/docstrings, JavaScript/TypeScript block/doc/line comments and
+  strings, HTML comments, and CSS block comments/strings.
+- Protected comment/string ranges are cached by buffer revision and mapped onto
+  viewport lines before `theme_bridge.tokens_to_spans()` flattens TextMate
+  output, so protected comment/string style wins over leaked keyword, number,
+  operator, tag, selector, property, or value categories.
+
+Real tests:
+
+- `tests/extensions/test_textmate_multiline_protection.py` adds direct
+  protected-range tests, TextMate-span rendering tests, and editor-facing
+  rendering tests for Python, JavaScript, TypeScript, HTML, and CSS multiline
+  fixtures.
+- Existing performance coverage remains in
+  `tests/extensions/test_textmate_render_performance.py` and
+  `tests/extensions/test_textmate_scroll_regression.py` using real repository
+  files including `Makefile`, `logs/freebsd-0.2.2-fail.log`,
+  `logs/pr-46-body.md`, and `scripts/build_pyinstaller_linux.py`.
+
+Log/artifact analysis:
+
+- The known real artifacts for large-file acceptance remain the repository
+  `Makefile`, `logs/freebsd-0.2.2-fail.log`, and `logs/pr-46-body.md`.
+- Synthetic fixtures are intentionally used only for exact adversarial
+  multiline comment/string bodies that are not guaranteed to exist in repository
+  files.
+
+Documentation:
+
+- `docs/architecture/extensions-layer.md` documents TextMate-primary rendering,
+  the bounded language-aware protection layer, and the large-file/multiline
+  acceptance tests.
+- `docs/release/release-checklist.md` includes large-file scroll, multiline
+  rendering, no-SQL fallback, and TextMate dependency/fallback checks.
+
+Audit conclusion:
+
+- Imported upstream extension assets remain untouched.
+- F11 PySH Console Panel behavior, future F4 linter work, and VMLab/QEMU/QMP
+  scope remain untouched.
+- The issue #102 rendering gap is constrained to the ECLI-owned adapter layer,
+  with focused regression and performance evidence.