docs/ci: Append COAMS Developer Guide chapter and patch CI matrices

Author: quantDIY

.github/workflows/coams-verification.yml

@@ -11,6 +11,9 @@ concurrency:
   group: ${{ github.workflow }}-${{ github.ref }}
   cancel-in-progress: true
 
+env:
+  FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: true
+
 jobs:
   # Pillar 1: Fuzzing the Ignorant Engine (Fail Fast)
   agent-index-fuzzing:
@@ -42,7 +45,7 @@ jobs:
         uses: actions/setup-go@v5
         with:
           go-version: '1.22'
-          cache: true
+          cache: false
 
       - name: Run Publish Saga Suite
         # Standard go test with race detector enabled for concurrent saga checks

docs/DEVELOPER_GUIDE.md

@@ -43,3 +43,26 @@ GERP utilizes mathematically reproducible cloud infrastructure and secure contai
 GERP exposes FAANG-grade capabilities autonomously to outside LLMs via standard JSON-RPC STDIO built off the Model Context Protocol.
 
 The `cmd/mcp` Server binds to AI agents running on Cursor IDE, Claude Desktop, or centralized orchestration engines via `.cursor/mcp.json`. Agents inherently use the physical tools (`gerp_status`, `gerp_create_order`, `gerp_audit_view`) to mechanically command the Spanner databases and trigger automated Temporal subroutines strictly adhering to the `internal/cli` environmental scope bindings.
+
+## 8. COAMS (Content Operating and Management System)
+
+COAMS represents the eighth primary domain of GERP, specifically architected as a Markdown-Native, AI-First content platform. Traditional headless CMS patterns relying on complex graphical JSON/ASTs have been discarded in favor of raw Markdown, natively optimizing it for AI integration.
+
+### Core Architecture & The "Ignorant Engine"
+- **Isolated AlloyDB Sharding:** Unlike 
+the core Spanner nodes, COAMS stores vector embeddings via AlloyDB using PostgreSQL's `pgvector` paired with Declarative Partitioning (`PARTITION BY LIST`). Data is physically separated by `channel_id` (e.g., engineering, hr) ensuring Zero-Leak Architecture.
+- **Microsecond Graph Linking:** The `doc:uuid` pointers are validated mathematically before writing by the **Agent-Index**. The system outright rejects writes if an internal link points to a non-existent or unauthorized document.
+
+### System Integration & The Publish Saga
+The COAMS ingestion logic lives entirely within the temporal queues (`internal/pipeline/coams_publish_saga.go`). Executing `gerp coams sync ./docs` triggers:
+1. Pure Go Goldmark AST markdown parsing.
+2. Link Integrity Verification.
+3. Vertex AI Vector Embeddings generation.
+4. Partitioned AlloyDB storage.
+5. Dynamic broadcast update of the GraphQL schema ensuring immediate downstream synchronization.
+
+### Bootstrapping the QuanuX Knowledge Vector
+By design, COAMS is self-documenting. To understand the operational footprint of the system, any AI agent or Operator can directly consult the embedded `SKILL.md` file:
+[Consult the COAMS SKILL.md Here](file:///Users/Duncan/GERP/internal/coams/docs/SKILL.md)
+
+Additionally, COAMS autonomously generates manual pages for CLI execution utilizing the `gerp coams gen-man` control hook. This fundamentally ensures that agents never suffer from blank-page syndrome when learning how to navigate the repository layer.

internal/coams/repository_integration_test.go

@@ -3,7 +3,6 @@
 package coams_test
 
 import (
-	"context"
 	"os"
 	"testing"
 )