@warlock.js/context 5.24.0 → 5.25.0

package/llms-full.txt

@@ -6,331 +6,331 @@
 
 ## define-context  `@warlock.js/context/define-context/SKILL.md`
 
----
-name: define-context
-description: 'Extend Context<TStore> to define an AsyncLocalStorage-backed typed context — implement buildStore, use run / enter / update / get / set / getStore / clear / hasContext. Triggers: `Context`, `Context<TStore>`, `buildStore`, `run`, `enter`, `update`, `get`, `set`, `getStore`, `clear`, `hasContext`; "share user/tenant/trace id across async calls", "AsyncLocalStorage typed wrapper", "request-scoped store without thread-through"; typical import `import { Context } from "@warlock.js/context"`. Skip: orchestrating multiple contexts — `@warlock.js/context/orchestrate-contexts/SKILL.md`; native `AsyncLocalStorage`, `cls-hooked`, `nest-context`, React Context.'
----
-
-# Define a context
-
-`@warlock.js/context` is a tiny wrapper on Node.js's `AsyncLocalStorage`. You extend the abstract `Context<TStore>` class to declare what your context stores, and you get a typed get/set/run API that propagates through every async call inside the scope.
-
-## Install
-
-```bash
-npm install @warlock.js/context
-```
-
-## Shape
-
-```ts
-import { Context } from "@warlock.js/context";
-
-interface UserContextStore {
-  userId: string;
-  role: "admin" | "user";
-  tenantId: string;
-}
-
-class UserContext extends Context<UserContextStore> {
-  /**
-   * Called by `contextManager.buildStores(payload)` for each registered context.
-   * Override to provide initialization logic for this context's store.
-   */
-  public buildStore(payload?: Record<string, any>): UserContextStore {
-    return {
-      userId: payload?.userId ?? "",
-      role: payload?.role ?? "user",
-      tenantId: payload?.tenantId ?? "",
-    };
-  }
-}
-
-export const userContext = new UserContext();
-```
-
-`buildStore` is the **only** abstract method. Everything else is provided by `Context<TStore>`.
-
-## Usage modes
-
-### `run()` — scoped execution
-
-```ts
-await userContext.run(
-  { userId: "123", role: "admin", tenantId: "acme" },
-  async () => {
-    // Context is available throughout this async scope and any awaited calls inside it.
-    const userId = userContext.get("userId");   // "123"
-    const role = userContext.get("role");       // "admin"
-
-    await someAsyncOperation();                  // context propagates through awaits
-  },
-);
-```
-
-Use `run()` when you have a clear scope boundary (a request handler, a job, a CLI command). The context is auto-cleaned when the callback returns.
-
-### `enter()` — middleware-style, no callback
-
-```ts
-function authMiddleware(req, res, next) {
-  userContext.enter({
-    userId: req.user.id,
-    role: req.user.role,
-    tenantId: req.headers["x-tenant-id"],
-  });
-
-  next();   // context lives for the rest of the request
-}
-```
-
-Use `enter()` when the framework doesn't give you a callback to wrap (Express-style middleware). Under the hood it's `AsyncLocalStorage.enterWith(store)`.
-
-### `update()` — merge into the current context
-
-```ts
-userContext.update({ role: "admin" });
-// existing store: { userId, role, tenantId } → { userId, role: "admin", tenantId }
-```
-
-If there's no current store, `update` creates one with the partial (cast to the full type). Use for incremental enrichment as the request flows through layers.
-
-`update()` (and `set()`, which calls it) silently drops `__proto__` / `constructor` / `prototype` keys instead of merging them — merging with `Object.assign` would otherwise let a caller-shaped payload (e.g. `userContext.update(req.body)`) pollute `Object.prototype` for the whole process, since the context store is shared. The rest of the payload still merges normally; this closes only the prototype-pollution vector, not general trust of merged input — validate before you merge caller-supplied data.
-
-## Reading
-
-```ts
-const userId = userContext.get("userId");         // TStore[K] | undefined
-const store = userContext.getStore();              // TStore | undefined
-const inside = userContext.hasContext();           // boolean
-```
-
-`get` is the daily-use accessor. `getStore` returns the whole record. `hasContext` distinguishes "key absent" from "no context at all" (which matters for safety checks at the framework boundary).
-
-## Writing within a context
-
-```ts
-userContext.set("role", "admin");                  // sugar for update({ role: "admin" })
-```
-
-Only call `set` inside an active context. Outside one, it enters a new context with just that key set — usually not what you want.
-
-## Clearing
-
-```ts
-userContext.clear();
-// replaces the current store with an empty object of TStore
-```
-
-Rare in app code. The auto-cleanup at the end of `run()` is the normal path.
-
-## Convenience getters via subclass
-
-Add domain-friendly getters on the subclass when a key is read a lot:
-
-```ts
-class TenantContext extends Context<TenantStore> {
-  public buildStore(payload?: Record<string, any>): TenantStore {
-    return {
-      tenantId: payload?.tenantId ?? "",
-      tenantName: payload?.tenantName ?? "",
-      config: payload?.config ?? {},
-    };
-  }
-
-  public get tenantId() {
-    return this.get("tenantId");
-  }
-
-  public get config() {
-    return this.get("config");
-  }
-}
-```
-
-Now `tenantContext.tenantId` reads better than `tenantContext.get("tenantId")` at the call site.
-
-## What it's NOT for
-
-- **Persistent state across requests.** AsyncLocalStorage is per-call-tree; data dies when the scope ends. Use a cache, a database, or a singleton for cross-request data.
-- **Thread-safe shared mutable state.** Each `run()` gets a fresh store. Two parallel `run()` calls don't see each other's updates.
-- **Sync code that doesn't `await` anything.** Works, but if there's no async boundary the context add-overhead is wasted — just pass the data as a parameter.
-
-## See also
-
-- [`@warlock.js/context/orchestrate-contexts/SKILL.md`](@warlock.js/context/orchestrate-contexts/SKILL.md) — running multiple contexts together via the `contextManager` singleton.
-
-## Things NOT to do
-
-- Don't make every cross-cutting concern a context. Build one when it has its own lifecycle (request, db transaction, trace span). For one-off data, a function argument is clearer.
-- Don't capture the store reference outside the scope — it's freed when `run()` ends. Read the value out before exiting the scope if you need it later.
-- Don't mutate the store object directly without `update`/`set` — works, but obscures intent. The methods exist to make state changes searchable.
-- Don't share one context instance across unrelated concerns. One typed context per domain concept reads better than one fat `globalContext`.
+---
+name: define-context
+description: 'Extend Context<TStore> to define an AsyncLocalStorage-backed typed context — implement buildStore, use run / enter / update / get / set / getStore / clear / hasContext. Triggers: `Context`, `Context<TStore>`, `buildStore`, `run`, `enter`, `update`, `get`, `set`, `getStore`, `clear`, `hasContext`; "share user/tenant/trace id across async calls", "AsyncLocalStorage typed wrapper", "request-scoped store without thread-through"; typical import `import { Context } from "@warlock.js/context"`. Skip: orchestrating multiple contexts — `@warlock.js/context/orchestrate-contexts/SKILL.md`; native `AsyncLocalStorage`, `cls-hooked`, `nest-context`, React Context.'
+---
+
+# Define a context
+
+`@warlock.js/context` is a tiny wrapper on Node.js's `AsyncLocalStorage`. You extend the abstract `Context<TStore>` class to declare what your context stores, and you get a typed get/set/run API that propagates through every async call inside the scope.
+
+## Install
+
+```bash
+npm install @warlock.js/context
+```
+
+## Shape
+
+```ts
+import { Context } from "@warlock.js/context";
+
+interface UserContextStore {
+  userId: string;
+  role: "admin" | "user";
+  tenantId: string;
+}
+
+class UserContext extends Context<UserContextStore> {
+  /**
+   * Called by `contextManager.buildStores(payload)` for each registered context.
+   * Override to provide initialization logic for this context's store.
+   */
+  public buildStore(payload?: Record<string, any>): UserContextStore {
+    return {
+      userId: payload?.userId ?? "",
+      role: payload?.role ?? "user",
+      tenantId: payload?.tenantId ?? "",
+    };
+  }
+}
+
+export const userContext = new UserContext();
+```
+
+`buildStore` is the **only** abstract method. Everything else is provided by `Context<TStore>`.
+
+## Usage modes
+
+### `run()` — scoped execution
+
+```ts
+await userContext.run(
+  { userId: "123", role: "admin", tenantId: "acme" },
+  async () => {
+    // Context is available throughout this async scope and any awaited calls inside it.
+    const userId = userContext.get("userId");   // "123"
+    const role = userContext.get("role");       // "admin"
+
+    await someAsyncOperation();                  // context propagates through awaits
+  },
+);
+```
+
+Use `run()` when you have a clear scope boundary (a request handler, a job, a CLI command). The context is auto-cleaned when the callback returns.
+
+### `enter()` — middleware-style, no callback
+
+```ts
+function authMiddleware(req, res, next) {
+  userContext.enter({
+    userId: req.user.id,
+    role: req.user.role,
+    tenantId: req.headers["x-tenant-id"],
+  });
+
+  next();   // context lives for the rest of the request
+}
+```
+
+Use `enter()` when the framework doesn't give you a callback to wrap (Express-style middleware). Under the hood it's `AsyncLocalStorage.enterWith(store)`.
+
+### `update()` — merge into the current context
+
+```ts
+userContext.update({ role: "admin" });
+// existing store: { userId, role, tenantId } → { userId, role: "admin", tenantId }
+```
+
+If there's no current store, `update` creates one with the partial (cast to the full type). Use for incremental enrichment as the request flows through layers.
+
+`update()` (and `set()`, which calls it) silently drops `__proto__` / `constructor` / `prototype` keys instead of merging them — merging with `Object.assign` would otherwise let a caller-shaped payload (e.g. `userContext.update(req.body)`) pollute `Object.prototype` for the whole process, since the context store is shared. The rest of the payload still merges normally; this closes only the prototype-pollution vector, not general trust of merged input — validate before you merge caller-supplied data.
+
+## Reading
+
+```ts
+const userId = userContext.get("userId");         // TStore[K] | undefined
+const store = userContext.getStore();              // TStore | undefined
+const inside = userContext.hasContext();           // boolean
+```
+
+`get` is the daily-use accessor. `getStore` returns the whole record. `hasContext` distinguishes "key absent" from "no context at all" (which matters for safety checks at the framework boundary).
+
+## Writing within a context
+
+```ts
+userContext.set("role", "admin");                  // sugar for update({ role: "admin" })
+```
+
+Only call `set` inside an active context. Outside one, it enters a new context with just that key set — usually not what you want.
+
+## Clearing
+
+```ts
+userContext.clear();
+// replaces the current store with an empty object of TStore
+```
+
+Rare in app code. The auto-cleanup at the end of `run()` is the normal path.
+
+## Convenience getters via subclass
+
+Add domain-friendly getters on the subclass when a key is read a lot:
+
+```ts
+class TenantContext extends Context<TenantStore> {
+  public buildStore(payload?: Record<string, any>): TenantStore {
+    return {
+      tenantId: payload?.tenantId ?? "",
+      tenantName: payload?.tenantName ?? "",
+      config: payload?.config ?? {},
+    };
+  }
+
+  public get tenantId() {
+    return this.get("tenantId");
+  }
+
+  public get config() {
+    return this.get("config");
+  }
+}
+```
+
+Now `tenantContext.tenantId` reads better than `tenantContext.get("tenantId")` at the call site.
+
+## What it's NOT for
+
+- **Persistent state across requests.** AsyncLocalStorage is per-call-tree; data dies when the scope ends. Use a cache, a database, or a singleton for cross-request data.
+- **Thread-safe shared mutable state.** Each `run()` gets a fresh store. Two parallel `run()` calls don't see each other's updates.
+- **Sync code that doesn't `await` anything.** Works, but if there's no async boundary the context add-overhead is wasted — just pass the data as a parameter.
+
+## See also
+
+- [`@warlock.js/context/orchestrate-contexts/SKILL.md`](@warlock.js/context/orchestrate-contexts/SKILL.md) — running multiple contexts together via the `contextManager` singleton.
+
+## Things NOT to do
+
+- Don't make every cross-cutting concern a context. Build one when it has its own lifecycle (request, db transaction, trace span). For one-off data, a function argument is clearer.
+- Don't capture the store reference outside the scope — it's freed when `run()` ends. Read the value out before exiting the scope if you need it later.
+- Don't mutate the store object directly without `update`/`set` — works, but obscures intent. The methods exist to make state changes searchable.
+- Don't share one context instance across unrelated concerns. One typed context per domain concept reads better than one fat `globalContext`.
 
 
 ## orchestrate-contexts  `@warlock.js/context/orchestrate-contexts/SKILL.md`
 
----
-name: orchestrate-contexts
-description: 'Orchestrate multiple Context<TStore> instances via the contextManager singleton — register, buildStores, runAll, enterAll, clearAll. Triggers: `contextManager`, `register`, `buildStores`, `runAll`, `enterAll`, `clearAll`, `unregister`, `getContext`, `hasContext`; "run multiple contexts active for the same scope", "register contexts at boot", "avoid nested run() calls for request + database + tenant"; typical import `import { contextManager } from "@warlock.js/context"`. Skip: defining a single context class — `@warlock.js/context/define-context/SKILL.md`; native `AsyncLocalStorage` nesting, `cls-hooked` namespaces.'
----
-
-# Orchestrate contexts — `contextManager`
-
-`contextManager` is a singleton that knows about every registered `Context` and runs them all together so you don't write nested `run()` calls by hand.
-
-## Why use it
-
-You can call `context.run(store, fn)` directly when you only have one context. With two or more, the manager handles the nesting:
-
-```ts
-// ❌ Without the manager — fragile, easy to forget a layer
-await requestContext.run(reqStore, async () =>
-  databaseContext.run(dbStore, async () =>
-    tenantContext.run(tenantStore, async () => handle()),
-  ),
-);
-
-// ✅ With the manager — one call, deterministic order
-await contextManager.runAll({ request: reqStore, database: dbStore, tenant: tenantStore }, handle);
-```
-
-## Register contexts at boot
-
-```ts
-import { contextManager } from "@warlock.js/context";
-import { requestContext } from "./request-context";
-import { databaseContext } from "./database-context";
-import { tenantContext } from "./tenant-context";
-
-contextManager
-  .register("request", requestContext)
-  .register("database", databaseContext)
-  .register("tenant", tenantContext);
-```
-
-Returns the manager — chain registrations. Names must be unique; re-registering with the same name overwrites.
-
-## Build stores + run
-
-The typical flow is two-step: build initial stores from a request-like payload, then run.
-
-```ts
-app.use(async (req, res, next) => {
-  const stores = contextManager.buildStores({
-    request: req,
-    response: res,
-    tenantId: req.headers["x-tenant-id"],
-  });
-
-  await contextManager.runAll(stores, async () => {
-    await next();
-  });
-});
-```
-
-`buildStores(payload)` calls each registered context's `buildStore(payload)` (the abstract method on `Context<TStore>` — see [`@warlock.js/context/define-context/SKILL.md`](@warlock.js/context/define-context/SKILL.md)) and returns `{ [contextName]: store }`.
-
-`runAll(stores, fn)` nests every context's `run()` in registration order, then invokes `fn` at the innermost layer. All contexts are active inside `fn`.
-
-## `enterAll()` — middleware without a callback
-
-When the framework doesn't give you a callback to wrap (e.g. Express middleware where you call `next()` and return):
-
-```ts
-function contextMiddleware(req, res, next) {
-  const stores = contextManager.buildStores({ request: req, response: res });
-  contextManager.enterAll(stores);
-  next();
-}
-```
-
-`enterAll` calls `enter()` on each registered context whose name has a **truthy** store value — a `name` with no key (or a falsy value like `undefined`) is skipped, leaving any already-active store for that context untouched. The entered contexts live for the rest of the request. Each can still be `clear()`-ed later. Use `runAll` when you can — `enterAll` doesn't auto-clean.
-
-## Lookup + introspection
-
-```ts
-contextManager.hasContext("tenant");                // boolean
-const tenant = contextManager.getContext<TenantContext>("tenant");
-// returns the registered instance or undefined
-contextManager.unregister("debug");                  // remove a context
-contextManager.clearAll();                           // clear stores on every context
-```
-
-`getContext<T>` returns the registered instance cast to `T` (the generic is constrained to `Context<any>`, so pass the concrete context class — `getContext<TenantContext>("tenant")`). It's an unchecked cast: an unknown `name` returns `undefined`, and a wrong type argument won't be caught at runtime. Useful in shared utilities that want to read from a context without importing it at the call site.
-
-## Order matters
-
-`runAll` nests in **registration order**. The first-registered context is the outermost layer. If contexts have ordering constraints (database before tenant because tenant resolves via the db), register them in that order.
-
-```ts
-contextManager
-  .register("trace", traceContext)        // outermost — runs first
-  .register("request", requestContext)
-  .register("database", databaseContext)
-  .register("tenant", tenantContext);     // innermost — runs last, inside all others
-```
-
-## Real-world: multi-tenant request lifecycle
-
-```ts
-import { Context, contextManager } from "@warlock.js/context";
-import { randomUUID } from "crypto";
-
-class TraceContext extends Context<{ traceId: string; startTime: number }> {
-  public buildStore(): { traceId: string; startTime: number } {
-    return { traceId: randomUUID(), startTime: Date.now() };
-  }
-
-  public get traceId() {
-    return this.get("traceId");
-  }
-}
-
-class RequestContext extends Context<{ request: any; response: any }> {
-  public buildStore(payload?: any) {
-    return { request: payload?.request, response: payload?.response };
-  }
-}
-
-class TenantContext extends Context<{ tenantId: string }> {
-  public buildStore(payload?: any) {
-    return { tenantId: payload?.tenantId ?? "" };
-  }
-}
-
-export const traceContext = new TraceContext();
-export const requestContext = new RequestContext();
-export const tenantContext = new TenantContext();
-
-contextManager
-  .register("trace", traceContext)
-  .register("request", requestContext)
-  .register("tenant", tenantContext);
-
-// In your HTTP layer:
-async function handleRequest(req: any, res: any) {
-  const stores = contextManager.buildStores({
-    request: req,
-    response: res,
-    tenantId: req.headers["x-tenant-id"],
-  });
-
-  return contextManager.runAll(stores, async () => {
-    // All three contexts active here:
-    console.log(`Trace ${traceContext.get("traceId")} — tenant ${tenantContext.get("tenantId")}`);
-
-    await routeAndDispatch(req, res);
-  });
-}
-```
-
-## Things NOT to do
-
-- Don't register the same context under two names — every context is its own singleton and shares its store across registrations, but multiple names confuse `buildStores` (the payload is split per name).
-- Don't `runAll` an empty stores map — every key without a registered context is silently ignored, and missing contexts get `{}` as their store. Better to construct the stores explicitly and fail loudly when a key is missing.
-- Don't use the manager when only one context applies. `context.run(store, fn)` is shorter and has the same semantics.
-- Don't expect `enterAll()` to auto-clean. It's a one-way setup — pair with `clearAll()` at the end of the request, or just use `runAll` when you can.
+---
+name: orchestrate-contexts
+description: 'Orchestrate multiple Context<TStore> instances via the contextManager singleton — register, buildStores, runAll, enterAll, clearAll. Triggers: `contextManager`, `register`, `buildStores`, `runAll`, `enterAll`, `clearAll`, `unregister`, `getContext`, `hasContext`; "run multiple contexts active for the same scope", "register contexts at boot", "avoid nested run() calls for request + database + tenant"; typical import `import { contextManager } from "@warlock.js/context"`. Skip: defining a single context class — `@warlock.js/context/define-context/SKILL.md`; native `AsyncLocalStorage` nesting, `cls-hooked` namespaces.'
+---
+
+# Orchestrate contexts — `contextManager`
+
+`contextManager` is a singleton that knows about every registered `Context` and runs them all together so you don't write nested `run()` calls by hand.
+
+## Why use it
+
+You can call `context.run(store, fn)` directly when you only have one context. With two or more, the manager handles the nesting:
+
+```ts
+// ❌ Without the manager — fragile, easy to forget a layer
+await requestContext.run(reqStore, async () =>
+  databaseContext.run(dbStore, async () =>
+    tenantContext.run(tenantStore, async () => handle()),
+  ),
+);
+
+// ✅ With the manager — one call, deterministic order
+await contextManager.runAll({ request: reqStore, database: dbStore, tenant: tenantStore }, handle);
+```
+
+## Register contexts at boot
+
+```ts
+import { contextManager } from "@warlock.js/context";
+import { requestContext } from "./request-context";
+import { databaseContext } from "./database-context";
+import { tenantContext } from "./tenant-context";
+
+contextManager
+  .register("request", requestContext)
+  .register("database", databaseContext)
+  .register("tenant", tenantContext);
+```
+
+Returns the manager — chain registrations. Names must be unique; re-registering with the same name overwrites.
+
+## Build stores + run
+
+The typical flow is two-step: build initial stores from a request-like payload, then run.
+
+```ts
+app.use(async (req, res, next) => {
+  const stores = contextManager.buildStores({
+    request: req,
+    response: res,
+    tenantId: req.headers["x-tenant-id"],
+  });
+
+  await contextManager.runAll(stores, async () => {
+    await next();
+  });
+});
+```
+
+`buildStores(payload)` calls each registered context's `buildStore(payload)` (the abstract method on `Context<TStore>` — see [`@warlock.js/context/define-context/SKILL.md`](@warlock.js/context/define-context/SKILL.md)) and returns `{ [contextName]: store }`.
+
+`runAll(stores, fn)` nests every context's `run()` in registration order, then invokes `fn` at the innermost layer. All contexts are active inside `fn`.
+
+## `enterAll()` — middleware without a callback
+
+When the framework doesn't give you a callback to wrap (e.g. Express middleware where you call `next()` and return):
+
+```ts
+function contextMiddleware(req, res, next) {
+  const stores = contextManager.buildStores({ request: req, response: res });
+  contextManager.enterAll(stores);
+  next();
+}
+```
+
+`enterAll` calls `enter()` on each registered context whose name has a **truthy** store value — a `name` with no key (or a falsy value like `undefined`) is skipped, leaving any already-active store for that context untouched. The entered contexts live for the rest of the request. Each can still be `clear()`-ed later. Use `runAll` when you can — `enterAll` doesn't auto-clean.
+
+## Lookup + introspection
+
+```ts
+contextManager.hasContext("tenant");                // boolean
+const tenant = contextManager.getContext<TenantContext>("tenant");
+// returns the registered instance or undefined
+contextManager.unregister("debug");                  // remove a context
+contextManager.clearAll();                           // clear stores on every context
+```
+
+`getContext<T>` returns the registered instance cast to `T` (the generic is constrained to `Context<any>`, so pass the concrete context class — `getContext<TenantContext>("tenant")`). It's an unchecked cast: an unknown `name` returns `undefined`, and a wrong type argument won't be caught at runtime. Useful in shared utilities that want to read from a context without importing it at the call site.
+
+## Order matters
+
+`runAll` nests in **registration order**. The first-registered context is the outermost layer. If contexts have ordering constraints (database before tenant because tenant resolves via the db), register them in that order.
+
+```ts
+contextManager
+  .register("trace", traceContext)        // outermost — runs first
+  .register("request", requestContext)
+  .register("database", databaseContext)
+  .register("tenant", tenantContext);     // innermost — runs last, inside all others
+```
+
+## Real-world: multi-tenant request lifecycle
+
+```ts
+import { Context, contextManager } from "@warlock.js/context";
+import { randomUUID } from "crypto";
+
+class TraceContext extends Context<{ traceId: string; startTime: number }> {
+  public buildStore(): { traceId: string; startTime: number } {
+    return { traceId: randomUUID(), startTime: Date.now() };
+  }
+
+  public get traceId() {
+    return this.get("traceId");
+  }
+}
+
+class RequestContext extends Context<{ request: any; response: any }> {
+  public buildStore(payload?: any) {
+    return { request: payload?.request, response: payload?.response };
+  }
+}
+
+class TenantContext extends Context<{ tenantId: string }> {
+  public buildStore(payload?: any) {
+    return { tenantId: payload?.tenantId ?? "" };
+  }
+}
+
+export const traceContext = new TraceContext();
+export const requestContext = new RequestContext();
+export const tenantContext = new TenantContext();
+
+contextManager
+  .register("trace", traceContext)
+  .register("request", requestContext)
+  .register("tenant", tenantContext);
+
+// In your HTTP layer:
+async function handleRequest(req: any, res: any) {
+  const stores = contextManager.buildStores({
+    request: req,
+    response: res,
+    tenantId: req.headers["x-tenant-id"],
+  });
+
+  return contextManager.runAll(stores, async () => {
+    // All three contexts active here:
+    console.log(`Trace ${traceContext.get("traceId")} — tenant ${tenantContext.get("tenantId")}`);
+
+    await routeAndDispatch(req, res);
+  });
+}
+```
+
+## Things NOT to do
+
+- Don't register the same context under two names — every context is its own singleton and shares its store across registrations, but multiple names confuse `buildStores` (the payload is split per name).
+- Don't `runAll` an empty stores map — every key without a registered context is silently ignored, and missing contexts get `{}` as their store. Better to construct the stores explicitly and fail loudly when a key is missing.
+- Don't use the manager when only one context applies. `context.run(store, fn)` is shorter and has the same semantics.
+- Don't expect `enterAll()` to auto-clean. It's a one-way setup — pair with `clearAll()` at the end of the request, or just use `runAll` when you can.
 
 
 ## overview  `@warlock.js/context/overview/SKILL.md`