✨ feat(context-preview): in-context review panel with segment-aware highlighting

.gitignore

@@ -9,7 +9,10 @@ nbproject/
 webgrind/
 _dic_cache
 .phpunit.result.cache
-/docs
+/docs/*
+!/docs/*.md
+!/docs/*.png
+!/docs/reviews/
 
 # Configuration files
 .num_processes

CLAUDE.md

@@ -0,0 +1,50 @@
+# Memory-First Protocol
+<!-- signet-first-version: 1.2.0 -->
+
+These rules enforce memory-aware behavior for AI coding agents.
+If `signet_memory_search` is available, use Signet as the primary memory system.
+Otherwise, use your native memory capabilities (MEMORY.md, auto memory, etc.).
+The principles below apply regardless of which memory backend is available.
+
+## 1. Search Memory Before Acting
+
+- ALWAYS search memory BEFORE reading files, firing agents, or running commands.
+- This includes session start: search for recent session summaries to orient before touching files.
+- **Certainty is the trigger, not doubt.** If you think "I already know this," that is EXACTLY when you must search.
+- Preferred: `signet_memory_search(query, type, limit)` when available.
+- Fallback: read MEMORY.md or use native recall.
+
+## 2. Pre-Action Gate
+
+- Before running ANY build, test, lint, deploy, or project-specific command, search memory for the verified procedure.
+- Use the stored version exactly. Do not add/remove flags from recall.
+- If not found in memory, check project files (Makefile, package.json, CI config). If still not found, ask the user.
+- NEVER skip to execution because you "remember" the command.
+
+## 3. Fallback Warning
+
+When memory search returns insufficient results and you fall back to reading files, inform the user:
+
+```
+MEMORY FALLBACK: Memory returned insufficient results for "<query>".
+Falling back to project files. Consider storing the result for future sessions.
+```
+
+This keeps the user aware of memory gaps so they can be filled over time.
+
+## 4. Store Conclusions
+
+- After any investigation, analysis, decision, debugging finding, or discovery, store the conclusion in memory.
+- Store synthesized conclusions, NOT raw data or transcripts.
+- Preferred: `signet_memory_store(content, type, tags, importance)` when available.
+- Fallback: use your native memory system.
+- Always check for duplicates before storing. Update existing entries rather than creating contradictions.
+
+## 5. Session Lifecycle
+
+- **Start**: Search memory for session summaries and recent context before touching files.
+- **End**: Store a summary of what was accomplished, decisions made, and unfinished work.
+
+---
+<!-- Do not edit above this line -- managed by signet-first plugin -->
+<!-- Add your project-specific rules below -->

INSTALL/matecat.sql

@@ -1126,7 +1126,7 @@ CREATE TABLE `segment_metadata` (
   `id_segment` bigint(20) NOT NULL,
   `meta_key` varchar(45) NOT NULL,
   `meta_value` varchar(255) NOT NULL,
-  KEY `idx_id_segment_meta_key` (`id_segment`,`meta_key`)
+  UNIQUE KEY `idx_id_segment_meta_key` (`id_segment`,`meta_key`)
 ) ENGINE=InnoDB DEFAULT CHARSET=utf8;
 /*!40101 SET character_set_client = @saved_cs_client */;
 

composer.json

@@ -10,14 +10,15 @@
         "defuse/php-encryption": "^2.4",
         "elasticsearch/elasticsearch": "8.*",
         "google/apiclient": "^2",
+        "google-gemini-php/client": "^2.7",
         "league/oauth2-facebook": "^2.2",
         "league/oauth2-github": "^3.1",
         "league/oauth2-linkedin": "^5.1",
         "matecat/cwh": "^3.1",
         "matecat/klein": "^3.2",
         "matecat/lara-php": "^v2.0.0-beta.1",
         "matecat/simple-s3": "^v2.0.0",
-        "matecat/subfiltering": "~v3.0",
+        "matecat/subfiltering": "~v4.0",
         "matecat/whole-text-finder": "^2.0",
         "matecat/xliff-parser": "~v3.0",
         "matecat/icu-intl": "^v1",
@@ -49,8 +50,7 @@
         "ext-openssl": "*",
         "ext-simplexml": "*",
         "ext-intl": "*",
-        "ext-redis": "*",
-        "google-gemini-php/client": "^2.7"
+        "ext-redis": "*"
     },
     "config": {
         "use-include-path": true,