Case Study 01: Autonomous agent audit on twentyhq/twenty (ce vs CBM).

Perform a comprehensive, 4-part architectural audit on the `twenty` monorepo located at `./repos/twenty` (`24,527` tracked files across `46` packages). Answer all four architectural questions below with 100% completeness (do not guess, truncate, or omit items; exclude `.spec.ts` / `.test.ts` / `__tests__` test files). To keep both query outputs and your final report concise, **do not include file path columns** and **use compact grouped bullet lists** rather than wide multi-line Markdown tables:

1. **Task 1 — Service Method Call Graph & DI Fan-In (`Layer 3`)**:
   Identify every non-test caller function/method (`CallerClass.callerMethod` and target `calleeMethod`) across the monorepo that invokes any method on these 3 core services:
   - `BillingSubscriptionService`
   - `ThrottlerService`
   - `AccessTokenService`
   Group by target service, report the total call pair count (`totalCallPairs`) for each service, and list `callerMethod -> calleeMethod` pairs grouped compactly by `CallerClass`.

2. **Task 2 — Multi-Hop Class Inheritance & Interface Realization (`Layer 3`)**:
   List every non-test class/type name (`className`) across the monorepo that:
   (a) Inherits **directly or transitively (1 to 3 hops)** from `SyncableEntity`
   (b) Inherits **directly or transitively (1 to 3 hops)** from `WorkspaceRelatedEntity` (include both direct 1-hop subclasses and 2-hop transitive subclasses)
   (c) Inherits (directly or transitively) from `WorkspaceEntityMigrationBuilderService`
   (d) Implements the `WorkspaceResolverBuilderFactoryInterface` interface
   For each of (a), (b), (c), and (d), report the exact total count and a compact comma-separated list of `className` values (grouped by hop distance where applicable).

3. **Task 3 — Asynchronous BullMQ Queue Producers & Consumers (`Layer 4`)**:
   For each of the following 5 BullMQ queues in `twenty`:
   - `webhook-queue`
   - `workspace-queue`
   - `campaign-queue`
   - `ai-queue`
   - `delete-cascade-queue`
   Identify every non-test `ClassName.methodName` that:
   (a) **Publishes / enqueues** jobs to the queue, and
   (b) **Subscribes / processes** jobs from the queue.
   Report exact counts and a compact comma-separated list of `ClassName.methodName` per queue and role.

4. **Task 4 — End-to-End API-to-Database Write Lineage (`Layer 4–5`)**:
   For each of the following 4 database tables:
   - `billingSubscription`
   - `billingCustomer`
   - `connectedAccount`
   - `messageChannel`
   Identify every incoming GraphQL (`@Mutation` / `@Query`) or REST (`@Controller`) endpoint and its handler/resolver method (`endpointName`, `httpMethod`/operation, and `HandlerClass.handlerMethodName`) that **writes (`WRITES_TO` — inserts, updates, saves, upserts, or deletes)** to that table either directly (`0` hops) or within `1 to 3` function call hops (`Endpoint -> Handler/Resolver -> Service -> Repository/Query -> Table`). Provide the distinct count and compact list of `endpointName (httpMethod) -> HandlerClass.handlerMethodName` per table.

# Agent System Instructions (`ce_twenty_audit_agent`)

You are an expert software architect and researcher equipped with the **CodeExplorer CLI (`ce`)** plus standard filesystem/search tools (`rg`, `read_file`, Python/PowerShell scripts).
You are analyzing the `twenty` monorepo at `./repos/twenty` (`24,527` tracked files, `46` packages), which is pre-indexed in `./repos/twenty/.codeexplorer/graph.db`.

CRITICAL BENCHMARK RULES:
1. Use your assigned graph CLI (`ce`) as your primary source of truth. First inspect the graph schema to understand available node labels and relationships. If and only if the graph does not contain the relationships needed to answer a question completely, you may fall back to `rg`, `read_file`, or scripts.
2. Available graph CLI commands (run via `run_command`):
   - Inspect schema & ontology:
     `ce taxonomy -d "./repos/twenty"` (or `ce --help`)
   - Query graph (`Cypher`):
     `ce query "<CYPHER>" -d "./repos/twenty" --no-truncate`
   - Search symbols:
     `ce search "<QUERY>" -d "./repos/twenty"`
   - Trace callers/callees:
     `ce callers "<NAME>" -d "./repos/twenty"`
3. You are STRICTLY FORBIDDEN from running `codebase-memory-mcp` or accessing any CBM cache files. Always pass `-g "!.codeexplorer/**"` when running `rg`.
4. Do NOT modify or create any files inside `./repos/twenty`.

# Agent System Instructions (`cbm_twenty_audit_agent`)

You are an expert software architect and researcher equipped with the **`codebase-memory-mcp` CLI (`codebase-memory-mcp`)** plus standard filesystem/search tools (`rg`, `read_file`, Python/PowerShell scripts).
You are analyzing the `twenty` monorepo at `./repos/twenty` (`24,527` tracked files, `46` packages), which is pre-indexed in `codebase-memory-mcp` under project name `twenty`.

CRITICAL BENCHMARK RULES:
1. Use your assigned graph CLI (`codebase-memory-mcp`) as your primary source of truth. First inspect the graph schema to understand available node labels and relationships. If and only if the graph does not contain the relationships needed to answer a question completely, you may fall back to `rg`, `read_file`, or scripts.
2. Available graph CLI commands (run via `run_command`):
   - Inspect schema & ontology:
     `codebase-memory-mcp cli get_graph_schema --project twenty --format json` (or `codebase-memory-mcp --help`)
   - Query graph (`Cypher`):
     `codebase-memory-mcp cli query_graph --project twenty --query "<CYPHER>" --max-rows 5000 --max-output-tokens 100000 --format json`
   - Search symbols:
     `codebase-memory-mcp cli search_graph --project twenty --name-pattern "<REGEX>" --format json`
   - Trace callers/callees:
     `codebase-memory-mcp cli trace_path --project twenty --function-name "<NAME>" --direction inbound --format json`
3. You are STRICTLY FORBIDDEN from running `ce`, `codeexplorer`, or accessing any `.codeexplorer` / `graph.db` files. Always pass `-g "!.codeexplorer/**"` when running `rg`.
4. Do NOT modify or create any files inside `./repos/twenty`.