hotmilk 0.1.8 → 0.1.10

package/AGENTS.md

@@ -6,14 +6,14 @@ User-facing setup: [README.md](README.md).
 
 ## Commands (this repo)
 
-| Command         | Purpose                          |
-| --------------- | -------------------------------- |
-| `bun install`   | Install deps (`bun.lock`)        |
-| `bun test`      | Vitest via vite-plus             |
-| `bun run lint`  | `vp lint`                        |
-| `bun run check` | lint + format + test             |
-
-Requires **Node.js 22+**. Targets **Pi 0.77** (`@earendil-works/pi-coding-agent` peers).
+| Command         | Purpose                   |
+| --------------- | ------------------------- |
+| `bun install`   | Install deps (`bun.lock`) |
+| `bun test`      | Vitest via vite-plus      |
+| `bun run lint`  | `vp lint`                 |
+| `bun run check` | lint + format + test      |
+
+Requires **Node.js 22+**. Targets **Pi 0.78** (`@earendil-works/pi-coding-agent` peers).
 
 ## Pi essentials
 

package/README.md

@@ -50,11 +50,11 @@ pi install -l npm:hotmilk
 2. On first session, hotmilk creates `~/.pi/agent/hotmilk.json` if missing (defaults match the bundled template).
 3. After config changes, run `/reload`.
 
-### Pi 0.77 and npm peers
+### Pi 0.78 and npm peers
 
-hotmilk targets **Pi 0.77** (`@earendil-works/pi-coding-agent` and peers). Several bundled dependencies still declare **0.74.x** peer ranges (`pi-simplify`, `pi-btw`, `pi-docparser`, `@blackbelt-technology/pi-flows`, and others). npm may report `ERESOLVE` until those packages publish 0.77-compatible peers.
+hotmilk targets **Pi 0.78** (`@earendil-works/pi-coding-agent` and peers). Several bundled dependencies still declare **0.74.x** peer ranges (`pi-simplify`, `pi-btw`, `pi-docparser`, `@blackbelt-technology/pi-flows`, and others). npm may report `ERESOLVE` until those packages publish 0.78-compatible peers.
 
-This repo ships **`.npmrc`** with `legacy-peer-deps=true` so `npm install` and `npm ci` succeed. Copy from `.npmrc.example` if you clone without `.npmrc`. Treat upstream extensions as **best-effort on 0.77** until their maintainers widen peer ranges.
+This repo ships **`.npmrc`** with `legacy-peer-deps=true` so `npm install` and `npm ci` succeed. Copy from `.npmrc.example` if you clone without `.npmrc`. Treat upstream extensions as **best-effort on 0.78** until their maintainers widen peer ranges.
 
 Heavy optional stacks (`agent-dashboard`, `pi-flows`) stay **off by default**; enable in `/mode` only when you need them and accept extra peer / startup cost.
 
@@ -98,7 +98,7 @@ Or ask Pi to use the planning-with-files skill. The extension maintains `task_pl
     "cursor-provider": true,
     "btw": true,
     "simplify": true,
-    "rtk-optimizer": true,
+    "rtk-optimizer": false,
     "mcp-adapter": false,
     "planning-with-files": false,
     "caveman": false,
@@ -121,20 +121,22 @@ Or ask Pi to use the planning-with-files skill. The extension maintains `task_pl
 }
 ```
 
-| Key / area                        | Behavior                                                                                                           |
-| --------------------------------- | ------------------------------------------------------------------------------------------------------------------ |
-| `extensions.*`                    | Set to `false` to skip registering that bundled extension                                                          |
-| `extensions.subagents`            | Default `false`. When `true`, imports pi-subagents (~10s). Use with `gentle-ai` for delegation                     |
-| `extensions.goal` … `mcp-adapter` | Integration / perf extensions (formerly always loaded via `pi.extensions`; now toggled like other bundled deps)    |
-| Enabled extensions                | Loaded **in parallel** on session start (faster than sequential import when many toggles are on)                   |
-| `graph.warnOnStale`               | Notify when `graphify-out/needs_update` exists                                                                     |
-| `graph.autoSuggestUpdate`         | Append `graphify update .` to that notification                                                                    |
-| `defaults.persona`                | Seeds `.pi/gentle-ai/persona.json` when missing (`gentleman` \| `neutral`)                                         |
-| `defaults.language`               | Appends a project language hint to the system prompt each turn                                                     |
-| `mcp.seedOnStart`                 | Copy `mcp.json` template into `~/.pi/agent/mcp.json` when missing (empty template; for pi-mcp-adapter)             |
-| `extensions.mcp-adapter`          | Default `false`. Enable only when you want MCP servers from `~/.pi/agent/mcp.json` (do not duplicate context-mode) |
-
-**MCP (default):** `context-mode` extension registers `ctx_*` tools via its built-in bridge. Do **not** run `context-mode` as an MCP server at the same time. Hotmilk removes a legacy `context-mode` entry from `~/.pi/agent/mcp.json` on session start when the extension is on and mcp-adapter is off.
+| Key / area                        | Behavior                                                                                                                                                                                        |
+| --------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| `extensions.*`                    | Set to `false` to skip registering that bundled extension                                                                                                                                       |
+| `extensions.subagents`            | Default `false`. When `true`, imports pi-subagents (~10s). Use with `gentle-ai` for delegation                                                                                                  |
+| `extensions.context-mode`         | Default `true`. Prefer `ctx_*` for large outputs (see project context-window rules)                                                                                                             |
+| `extensions.rtk-optimizer`        | Default `false`. Bash/read/grep output compaction; enable with `context-mode` for leftover shell output. Install [`rtk` CLI](https://github.com/rtk-ai/rtk) for command rewrite (`/rtk verify`) |
+| `extensions.goal` … `mcp-adapter` | Integration / perf extensions (formerly always loaded via `pi.extensions`; now toggled like other bundled deps)                                                                                 |
+| Enabled extensions                | Loaded **in parallel** on session start (faster than sequential import when many toggles are on)                                                                                                |
+| `graph.warnOnStale`               | Notify when `graphify-out/needs_update` exists                                                                                                                                                  |
+| `graph.autoSuggestUpdate`         | Append `graphify update .` to that notification                                                                                                                                                 |
+| `defaults.persona`                | Seeds `.pi/gentle-ai/persona.json` when missing (`gentleman` \| `neutral`)                                                                                                                      |
+| `defaults.language`               | Appends a project language hint to the system prompt each turn                                                                                                                                  |
+| `mcp.seedOnStart`                 | Copy `mcp.json` template into `~/.pi/agent/mcp.json` when missing (empty template; for pi-mcp-adapter)                                                                                          |
+| `extensions.mcp-adapter`          | Default `false`. Enable only when you want MCP servers from `~/.pi/agent/mcp.json` (do not duplicate context-mode)                                                                              |
+
+**MCP (default):** `context-mode` extension registers `ctx_*` via its built-in bridge (same module as [upstream `.pi/extensions/context-mode`](https://github.com/mksglu/context-mode/tree/main/.pi/extensions/context-mode), loaded from `build/adapters/pi/extension.js`). Hotmilk removes any `context-mode` server from `~/.pi/agent/mcp.json` when the extension is on. Enable `mcp-adapter` only for **other** MCP servers—not a second context-mode entry.
 
 ### Optional extensions (off by default)
 
@@ -144,9 +146,9 @@ Enable in `/mode` or set the key to `true` in `hotmilk.json`, then `/reload`.
 | ----------------- | ------------------------------------------------------------------------------------------------------------------ | --------------------------------------------------------------------------------------------------------------- |
 | `caveman`         | [pi-caveman](https://www.npmjs.com/package/pi-caveman)                                                             | Terse English (`/caveman`). Conflicts with `defaults.language: ja`                                              |
 | `red-green`       | [pi-red-green](https://www.npmjs.com/package/pi-red-green)                                                         | TDD via `/tdd`, `/tdd-status`. Config: `~/.pi/red-green/config.json`                                            |
-| `agent-dashboard` | [@blackbelt-technology/pi-agent-dashboard](https://www.npmjs.com/package/@blackbelt-technology/pi-agent-dashboard) | Warm-starts via `pi-dashboard start` (~30s). Peers **0.74**; test on 0.77 before relying on it. Node.js ≥ 22.18 |
+| `agent-dashboard` | [@blackbelt-technology/pi-agent-dashboard](https://www.npmjs.com/package/@blackbelt-technology/pi-agent-dashboard) | Warm-starts via `pi-dashboard start` (~30s). Peers **0.74**; test on 0.78 before relying on it. Node.js ≥ 22.18 |
 | `web-access`      | [pi-web-access](https://www.npmjs.com/package/pi-web-access)                                                       | `web_search`, fetch, GitHub clone, PDF/video. Optional keys: `~/.pi/web-search.json`                            |
-| `pi-flows`        | [@blackbelt-technology/pi-flows](https://www.npmjs.com/package/@blackbelt-technology/pi-flows)                     | YAML DAG workflows (`/flows`). Peers **0.74** + `@sinclair/typebox`; off by default on 0.77 stacks              |
+| `pi-flows`        | [@blackbelt-technology/pi-flows](https://www.npmjs.com/package/@blackbelt-technology/pi-flows)                     | YAML DAG workflows (`/flows`). Peers **0.74** + `@sinclair/typebox`; off by default on 0.78 stacks              |
 
 **Agent dashboard troubleshooting**
 
@@ -170,7 +172,7 @@ Older **tabako** users: replace `npm:tabako` with `npm:hotmilk`; let hotmilk see
 
 ## Development
 
-Requires **Node.js 22+** (or **Bun 1.3+**), **Bun** for installs in this repo, and Pi **0.77** peers in the environment.
+Requires **Node.js 22+** (or **Bun 1.3+**), **Bun** for installs in this repo, and Pi **0.78** peers in the environment.
 
 ```bash
 bun install       # commit bun.lock; peers resolved by Bun
@@ -186,7 +188,7 @@ bun run check     # lint + format + test
 On push to `main`, GitHub Actions runs **lint + test**, then a **`publish` job** (`needs: test`) when `package.json` version is **newer than npm**. No separate workflow or tag push is required to start publish.
 
 ```text
-push main → test → publish (bun publish) → git tag v<version>
+push main → test → publish (npm publish --provenance) → git tag v<version>
 ```
 
 Bump `version` in `package.json` before pushing to `main`.
@@ -197,20 +199,20 @@ Bump `version` in `package.json` before pushing to `main`.
 2. Scope: publish to **`hotmilk`** (or classic `publish` on the account)
 3. Repository → **Settings → Secrets → Actions** → name **`NPM_TOKEN`**
 
-CI runs `bun publish --access public` with `NPM_TOKEN` (same secret name as [npm’s CI/CD doc](https://docs.npmjs.com/using-private-packages-in-a-ci-cd-workflow)). Dependencies are **not** bundled into the tarball (`bundleDependencies` removed — npm rejected the 162 MB hard-linked bundle with `E415`).
+CI uses [npm’s CI/CD workflow](https://docs.npmjs.com/using-private-packages-in-a-ci-cd-workflow): `actions/setup-node` with `registry-url`, then `npm publish --provenance --access public` with **`NODE_AUTH_TOKEN: ${{ secrets.NPM_TOKEN }}`**. The secret is named `NPM_TOKEN`; `setup-node` reads **`NODE_AUTH_TOKEN`** for auth. Dependencies are **not** bundled into the tarball (`bundleDependencies` removed — npm rejected the 162 MB hard-linked bundle with `E415`).
 
-Trusted Publisher on npm can stay configured or be removed; CI no longer depends on it.
+Trusted Publisher on npm can stay configured or be removed; CI uses the token path above.
 
 GitHub Release is optional — npm publish does not require it.
 
-Local publish: run `npm login` once, **or** add the token to **`~/.npmrc`** (not the repo `.npmrc`):
+Local publish: `npm login` once, then `npm publish --access public`. Or add the token to **`~/.npmrc`** (not the repo `.npmrc`):
 
 ```bash
 echo "//registry.npmjs.org/:_authToken=YOUR_NPM_TOKEN" >> ~/.npmrc
-bun publish --access public
+npm publish --access public
 ```
 
-`bun publish` does not read the `NPM_TOKEN` environment variable by itself — it uses `~/.npmrc` auth (same as npm).
+`bun publish` also works locally if `~/.npmrc` has a token; it does not read the `NPM_TOKEN` environment variable by itself.
 
 ### Layout
 

package/hotmilk.json

@@ -13,7 +13,7 @@
     "cursor-provider": true,
     "btw": true,
     "simplify": true,
-    "rtk-optimizer": true,
+    "rtk-optimizer": false,
     "mcp-adapter": false,
     "planning-with-files": false,
     "caveman": false,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "hotmilk",
-  "version": "0.1.8",
+  "version": "0.1.10",
   "description": "hotmilk — Pi package bundling gentle-pi, context-mode, graphify, subagents, and bundled extension toggles via hotmilk.json",
   "keywords": [
     "hotmilk",
@@ -60,21 +60,24 @@
     "pi-web-access": "^0.10.7"
   },
   "devDependencies": {
-    "@earendil-works/pi-agent-core": "^0.77.0",
-    "@earendil-works/pi-ai": "^0.77.0",
-    "@earendil-works/pi-coding-agent": "^0.77.0",
-    "@earendil-works/pi-tui": "^0.77.0",
-    "@types/node": "^22.15.0",
+    "@earendil-works/pi-agent-core": "^0.78.0",
+    "@earendil-works/pi-ai": "^0.78.0",
+    "@earendil-works/pi-coding-agent": "^0.78.0",
+    "@earendil-works/pi-tui": "^0.78.0",
+    "@types/node": "^22.19.19",
     "typebox": "^1.1.39",
     "vite-plus": "^0.1.23"
   },
   "peerDependencies": {
-    "@earendil-works/pi-agent-core": "^0.77.0",
-    "@earendil-works/pi-ai": "^0.77.0",
-    "@earendil-works/pi-coding-agent": "^0.77.0",
-    "@earendil-works/pi-tui": "^0.77.0",
+    "@earendil-works/pi-agent-core": "^0.78.0",
+    "@earendil-works/pi-ai": "^0.78.0",
+    "@earendil-works/pi-coding-agent": "^0.78.0",
+    "@earendil-works/pi-tui": "^0.78.0",
     "typebox": "^1.1.39"
   },
+  "engines": {
+    "node": "22"
+  },
   "pi": {
     "extensions": [
       "./src/index.ts"
@@ -100,8 +103,5 @@
     "themes": [
       "./themes"
     ]
-  },
-  "engines": {
-    "node": "22"
   }
 }

package/src/bootstrap/context-stack.ts

@@ -0,0 +1,167 @@
+import { existsSync, mkdirSync, readFileSync, writeFileSync } from "node:fs";
+import { dirname } from "node:path";
+import { getAgentDir } from "@earendil-works/pi-coding-agent";
+import { pruneContextModeMcpServerFromAgentConfig } from "../config/mcp.ts";
+import type { HotmilkRuntime } from "../config/runtime.ts";
+import type { BundledExtensionId } from "../config/hotmilk.ts";
+
+const RTK_EXTENSION_DIR = "pi-rtk-optimizer";
+const RTK_CONFIG_FILENAME = "config.json";
+
+/**
+ * Register in this order so context-mode hooks run before rtk-optimizer compacts tool output.
+ * Loader paths match upstream npm `pi.extensions` (not `.pi/extensions/` shims in the git repo).
+ */
+export const CONTEXT_STACK_EXTENSION_IDS = [
+  "context-mode",
+  "rtk-optimizer",
+] as const satisfies readonly BundledExtensionId[];
+
+export function getRtkOptimizerConfigPath(): string {
+  return `${getAgentDir()}/extensions/${RTK_EXTENSION_DIR}/${RTK_CONFIG_FILENAME}`;
+}
+
+export function expectedRtkMode(contextModeEnabled: boolean): "suggest" | "rewrite" {
+  return contextModeEnabled ? "suggest" : "rewrite";
+}
+
+/** Defaults when hotmilk seeds pi-rtk-optimizer config for the first time. */
+export function buildHotmilkRtkConfig(contextModeEnabled: boolean): Record<string, unknown> {
+  return {
+    enabled: true,
+    mode: expectedRtkMode(contextModeEnabled),
+    guardWhenRtkMissing: true,
+    showRewriteNotifications: false,
+    outputCompaction: {
+      enabled: true,
+      stripAnsi: true,
+      readCompaction: { enabled: false },
+      truncate: { enabled: true, maxChars: 12_000 },
+      sourceCodeFilteringEnabled: false,
+      preserveExactSkillReads: true,
+      sourceCodeFiltering: "none",
+      smartTruncate: { enabled: false, maxLines: 220 },
+      aggregateTestOutput: true,
+      filterBuildOutput: true,
+      compactGitOutput: true,
+      aggregateLinterOutput: true,
+      groupSearchOutput: true,
+      trackSavings: true,
+    },
+  };
+}
+
+type RtkConfigRecord = Record<string, unknown> & {
+  mode?: string;
+  outputCompaction?: {
+    readCompaction?: { enabled?: boolean };
+  };
+};
+
+function writeRtkConfig(configPath: string, config: unknown): void {
+  writeFileSync(configPath, `${JSON.stringify(config, null, 2)}\n`, "utf8");
+}
+
+export function seedRtkConfigIfMissing(
+  contextModeEnabled: boolean,
+  configPath = getRtkOptimizerConfigPath(),
+): { seeded: boolean; path: string } {
+  if (existsSync(configPath)) {
+    return { seeded: false, path: configPath };
+  }
+
+  mkdirSync(dirname(configPath), { recursive: true });
+  writeRtkConfig(configPath, buildHotmilkRtkConfig(contextModeEnabled));
+  return { seeded: true, path: configPath };
+}
+
+/** Hotmilk-managed fields only — does not touch Pi auto-compaction (settings.json). */
+function alignRtkConfigWithContextMode(
+  config: RtkConfigRecord,
+  contextModeEnabled: boolean,
+): boolean {
+  let changed = false;
+  const mode = expectedRtkMode(contextModeEnabled);
+  if (config.mode !== mode) {
+    config.mode = mode;
+    changed = true;
+  }
+
+  if (contextModeEnabled && config.outputCompaction?.readCompaction?.enabled !== false) {
+    config.outputCompaction ??= {};
+    config.outputCompaction.readCompaction = { enabled: false };
+    changed = true;
+  }
+
+  return changed;
+}
+
+export function syncRtkConfigForContextStack(
+  contextModeEnabled: boolean,
+  rtkEnabled: boolean,
+  configPath = getRtkOptimizerConfigPath(),
+): { updated: boolean; seeded: boolean; path: string } {
+  if (!rtkEnabled) {
+    return { updated: false, seeded: false, path: configPath };
+  }
+
+  if (!existsSync(configPath)) {
+    seedRtkConfigIfMissing(contextModeEnabled, configPath);
+    return { updated: true, seeded: true, path: configPath };
+  }
+
+  const config = JSON.parse(readFileSync(configPath, "utf8")) as RtkConfigRecord;
+  const changed = alignRtkConfigWithContextMode(config, contextModeEnabled);
+  if (changed) {
+    writeRtkConfig(configPath, config);
+  }
+
+  return { updated: changed, seeded: false, path: configPath };
+}
+
+/** Run before bundled extensions register (rtk seed/sync when that toggle is on). */
+export function prepareContextStack(extensionToggles: Record<BundledExtensionId, boolean>): void {
+  if (extensionToggles["rtk-optimizer"]) {
+    syncRtkConfigForContextStack(extensionToggles["context-mode"], true);
+  }
+}
+
+const MCP_PRUNED_MESSAGE = (path: string): string =>
+  `Removed duplicate context-mode entry from ${path}. ctx_* tools use the extension bridge only.`;
+
+const MCP_ADAPTER_DUPLICATE_WARNING =
+  "context-mode runs on the extension bridge (ctx_*). Do not add a context-mode server to mcp.json — use mcp-adapter only for other MCP servers.";
+
+const RTK_SYNC_MESSAGE =
+  "Adjusted pi-rtk-optimizer for context-mode coexistence (mode/readCompaction). Pi auto-compaction unchanged.";
+
+/**
+ * Option A: context-mode extension owns ctx_* via built-in MCP bridge.
+ * Prune legacy `context-mode` MCP server entries whenever the extension is enabled.
+ */
+export function applyContextStackOnSessionStart(
+  runtime: HotmilkRuntime,
+  notify: (message: string, level: "info" | "warning") => void,
+): void {
+  const { extensionToggles } = runtime;
+
+  if (extensionToggles["rtk-optimizer"]) {
+    const sync = syncRtkConfigForContextStack(extensionToggles["context-mode"], true);
+    if (sync.updated && !sync.seeded) {
+      notify(RTK_SYNC_MESSAGE, "info");
+    }
+  }
+
+  if (!extensionToggles["context-mode"]) {
+    return;
+  }
+
+  const mcpPrune = pruneContextModeMcpServerFromAgentConfig();
+  if (mcpPrune.pruned) {
+    notify(MCP_PRUNED_MESSAGE(mcpPrune.path), "info");
+  }
+
+  if (extensionToggles["mcp-adapter"]) {
+    notify(MCP_ADAPTER_DUPLICATE_WARNING, "warning");
+  }
+}

package/src/bootstrap/defaults.ts

@@ -3,6 +3,16 @@ import { dirname, join } from "node:path";
 import type { ExtensionAPI } from "@earendil-works/pi-coding-agent";
 import type { ResolvedDefaults } from "../config/hotmilk.ts";
 
+export const CAVEMAN_JA_CONFLICT_MESSAGE =
+  "caveman is on while defaults.language is ja — caveman adds English terse rules that fight the Japanese language hint. Turn off caveman (/mode), clear defaults.language, or use /caveman off.";
+
+export function shouldWarnCavemanJaConflict(
+  cavemanEnabled: boolean,
+  language: string | undefined,
+): boolean {
+  return cavemanEnabled && language?.trim().toLowerCase() === "ja";
+}
+
 export function seedPersonaFromDefaults(cwd: string, defaults: ResolvedDefaults): void {
   const path = join(cwd, ".pi", "gentle-ai", "persona.json");
   if (existsSync(path)) {

package/src/bootstrap/extensions.ts

@@ -1,4 +1,6 @@
 import type { ExtensionAPI } from "@earendil-works/pi-coding-agent";
+import { CONTEXT_STACK_EXTENSION_IDS } from "./context-stack.ts";
+import { bundledImportUrl } from "./resolve-bundled.ts";
 import { BUNDLED_EXTENSION_IDS, type BundledExtensionId } from "../config/hotmilk.ts";
 
 type ExtensionFactory = (pi: ExtensionAPI) => void | Promise<void>;
@@ -6,45 +8,59 @@ type ExtensionFactory = (pi: ExtensionAPI) => void | Promise<void>;
 /** Bundled deps may type against @mariozechner/pi-coding-agent; hotmilk uses @earendil-works. */
 type ExtensionModule = { default: unknown };
 
+function loadBundled(relativePath: string): () => Promise<ExtensionModule> {
+  return () => import(bundledImportUrl(relativePath));
+}
+
 /**
  * Load bundled extensions on demand so disabled toggles do not pay import cost.
- * Registrations run in parallel — extensions must not depend on each other's init order.
+ * Context stack registers sequentially; other enabled extensions load in parallel.
  */
 const BUNDLED_EXTENSION_LOADERS: Record<BundledExtensionId, () => Promise<ExtensionModule>> = {
-  "skill-registry": () => import("../../node_modules/gentle-pi/extensions/skill-registry.ts"),
-  "sdd-init": () => import("../../node_modules/gentle-pi/extensions/sdd-init.ts"),
-  "gentle-ai": () => import("../../node_modules/gentle-pi/extensions/gentle-ai.ts"),
-  "context-mode": () => import("../../node_modules/context-mode/build/adapters/pi/extension.js"),
-  "ask-user": () => import("pi-ask-user"),
-  graphify: () => import("../../node_modules/graphify-pi/extensions/graphify.ts"),
-  subagents: () => import("../../node_modules/pi-subagents/src/extension/index.ts"),
-  goal: () => import("../../node_modules/pi-goal/.pi/extensions/pi-goal/index.ts"),
-  docparser: () => import("../../node_modules/pi-docparser/extensions/docparser/index.ts"),
-  obsidian: () => import("../../node_modules/@haispeed/pi-obsidian/extensions/obsidian-cli.ts"),
-  "cursor-provider": () => import("../../node_modules/@netandreus/pi-cursor-provider/index.ts"),
-  btw: () => import("../../node_modules/pi-btw/extensions/btw.ts"),
-  simplify: () => import("../../node_modules/pi-simplify/dist/index.js"),
-  "rtk-optimizer": () => import("../../node_modules/pi-rtk-optimizer/index.ts"),
-  "mcp-adapter": () => import("../../node_modules/pi-mcp-adapter/index.ts"),
-  "planning-with-files": () =>
-    import("../../node_modules/@tomxprime/planning-with-files/extensions/planning-with-files/index.ts"),
-  caveman: () => import("../../node_modules/pi-caveman/extensions/caveman.ts"),
-  "red-green": () => import("../../node_modules/pi-red-green/dist/index.js"),
-  "agent-dashboard": () =>
-    import("../../node_modules/@blackbelt-technology/pi-agent-dashboard/packages/extension/src/bridge.ts"),
-  "web-access": () => import("../../node_modules/pi-web-access/index.ts"),
-  "pi-flows": () => import("../../node_modules/@blackbelt-technology/pi-flows/extensions/index.ts"),
+  "skill-registry": loadBundled("gentle-pi/extensions/skill-registry.ts"),
+  "sdd-init": loadBundled("gentle-pi/extensions/sdd-init.ts"),
+  "gentle-ai": loadBundled("gentle-pi/extensions/gentle-ai.ts"),
+  "context-mode": loadBundled("context-mode/build/adapters/pi/extension.js"),
+  "ask-user": loadBundled("pi-ask-user/index.ts"),
+  graphify: loadBundled("graphify-pi/extensions/graphify.ts"),
+  subagents: loadBundled("pi-subagents/src/extension/index.ts"),
+  goal: loadBundled("pi-goal/.pi/extensions/pi-goal/index.ts"),
+  docparser: loadBundled("pi-docparser/extensions/docparser/index.ts"),
+  obsidian: loadBundled("@haispeed/pi-obsidian/extensions/obsidian-cli.ts"),
+  "cursor-provider": loadBundled("@netandreus/pi-cursor-provider/index.ts"),
+  btw: loadBundled("pi-btw/extensions/btw.ts"),
+  simplify: loadBundled("pi-simplify/dist/index.js"),
+  "rtk-optimizer": loadBundled("pi-rtk-optimizer/index.ts"),
+  "mcp-adapter": loadBundled("pi-mcp-adapter/index.ts"),
+  "planning-with-files": loadBundled(
+    "@tomxprime/planning-with-files/extensions/planning-with-files/index.ts",
+  ),
+  caveman: loadBundled("pi-caveman/extensions/caveman.ts"),
+  "red-green": loadBundled("pi-red-green/dist/index.js"),
+  "agent-dashboard": loadBundled(
+    "@blackbelt-technology/pi-agent-dashboard/packages/extension/src/bridge.ts",
+  ),
+  "web-access": loadBundled("pi-web-access/index.ts"),
+  "pi-flows": loadBundled("@blackbelt-technology/pi-flows/extensions/index.ts"),
 };
 
+async function registerOne(pi: ExtensionAPI, id: BundledExtensionId): Promise<void> {
+  const mod = await BUNDLED_EXTENSION_LOADERS[id]();
+  await (mod.default as ExtensionFactory)(pi);
+}
+
 export async function registerBundledExtensions(
   pi: ExtensionAPI,
   enabled: Record<BundledExtensionId, boolean>,
 ): Promise<void> {
-  const enabledIds = BUNDLED_EXTENSION_IDS.filter((id) => enabled[id]);
-  await Promise.all(
-    enabledIds.map(async (id) => {
-      const mod = await BUNDLED_EXTENSION_LOADERS[id]();
-      await (mod.default as ExtensionFactory)(pi);
-    }),
-  );
+  const enabledIds = new Set(BUNDLED_EXTENSION_IDS.filter((id) => enabled[id]));
+
+  for (const id of CONTEXT_STACK_EXTENSION_IDS) {
+    if (enabledIds.has(id)) {
+      await registerOne(pi, id);
+      enabledIds.delete(id);
+    }
+  }
+
+  await Promise.all([...enabledIds].map((id) => registerOne(pi, id)));
 }

package/src/bootstrap/resolve-bundled.ts

@@ -0,0 +1,55 @@
+import { existsSync } from "node:fs";
+import { basename, dirname, join } from "node:path";
+import { fileURLToPath, pathToFileURL } from "node:url";
+
+/** Split `pkg/subpath` — supports scoped packages (`@scope/name/...`). */
+export function parseBundledModulePath(relativePath: string): { pkgName: string; subpath: string } {
+  if (relativePath.startsWith("@")) {
+    const parts = relativePath.split("/");
+    return { pkgName: `${parts[0]}/${parts[1]}`, subpath: parts.slice(2).join("/") };
+  }
+  const slash = relativePath.indexOf("/");
+  if (slash === -1) {
+    return { pkgName: relativePath, subpath: "index.ts" };
+  }
+  return { pkgName: relativePath.slice(0, slash), subpath: relativePath.slice(slash + 1) };
+}
+
+/**
+ * Resolve a bundled dependency file whether npm nested it under hotmilk or hoisted
+ * it next to hotmilk (e.g. `~/.pi/npm/node_modules/context-mode`).
+ */
+export function resolveBundledModule(
+  relativePath: string,
+  fromModuleUrl = import.meta.url,
+): string {
+  const { pkgName, subpath } = parseBundledModulePath(relativePath);
+  let dir = dirname(fileURLToPath(fromModuleUrl));
+
+  while (true) {
+    const nested = join(dir, "node_modules", pkgName, subpath);
+    if (existsSync(nested)) {
+      return nested;
+    }
+
+    if (basename(dir) === "node_modules") {
+      const sibling = join(dir, pkgName, subpath);
+      if (existsSync(sibling)) {
+        return sibling;
+      }
+    }
+
+    const parent = dirname(dir);
+    if (parent === dir) {
+      break;
+    }
+    dir = parent;
+  }
+
+  throw new Error(`Cannot resolve bundled module "${relativePath}" from hotmilk`);
+}
+
+/** Dynamic import URL for a bundled extension entry (works with nested and hoisted installs). */
+export function bundledImportUrl(relativePath: string): string {
+  return pathToFileURL(resolveBundledModule(relativePath)).href;
+}

package/src/bootstrap/rtk-coexist.ts

@@ -0,0 +1,6 @@
+/** @deprecated Import from `./context-stack.ts` instead. */
+export {
+  buildHotmilkRtkConfig,
+  getRtkOptimizerConfigPath,
+  seedRtkConfigIfMissing,
+} from "./context-stack.ts";

package/src/bootstrap/session.ts

@@ -1,58 +1,52 @@
 import type { ExtensionAPI } from "@earendil-works/pi-coding-agent";
-import {
-  pruneContextModeMcpServerFromAgentConfig,
-  seedAgentMcpJsonIfMissing,
-} from "../config/mcp.ts";
+import { applyContextStackOnSessionStart } from "./context-stack.ts";
+import { seedAgentMcpJsonIfMissing } from "../config/mcp.ts";
 import { AGENT_HOTMILK_CONFIG_LABEL, seedHotmilkConfigIfMissing } from "../config/hotmilk.ts";
 import type { HotmilkRuntime } from "../config/runtime.ts";
 import { setupHotmilkFooter } from "../ui/footer.ts";
-import { HOTMILK_LOGO } from "../ui/logo.ts";
-import { seedPersonaFromDefaults } from "./defaults.ts";
+import {
+  CAVEMAN_JA_CONFLICT_MESSAGE,
+  seedPersonaFromDefaults,
+  shouldWarnCavemanJaConflict,
+} from "./defaults.ts";
 
 const HOTMILK_SEEDED_MESSAGE = `Created ${AGENT_HOTMILK_CONFIG_LABEL} (toggle bundled extensions with /mode, then /reload).`;
 const HOTMILK_PARSE_ERROR_MESSAGE = (path: string, error: string): string =>
   `Failed to parse ${path}: ${error}. Using default extension toggles.`;
 const MCP_SEEDED_MESSAGE = (path: string): string =>
   `Created ${path} from hotmilk MCP template (add servers for pi-mcp-adapter; context-mode uses the extension bridge).`;
-const MCP_PRUNED_CONTEXT_MODE_MESSAGE = (path: string): string =>
-  `Removed duplicate context-mode entry from ${path}. Use context-mode extension tools (ctx_*), not MCP adapter, unless mcp-adapter is enabled.`;
 
 export function registerSessionHandlers(pi: ExtensionAPI, runtime: HotmilkRuntime): void {
   const termProgram = process.env.TERM_PROGRAM ?? "none";
 
   pi.on("session_start", (_event, ctx) => {
+    const uiNotify = (message: string, level: "info" | "warning") => ctx.ui.notify(message, level);
+
     setupHotmilkFooter(ctx, termProgram);
 
-    ctx.ui.notify(`\`\`\`text\n${HOTMILK_LOGO}\n\`\`\``, "info");
     const hotmilkSeed = seedHotmilkConfigIfMissing();
     if (hotmilkSeed.seeded) {
-      ctx.ui.notify(HOTMILK_SEEDED_MESSAGE, "info");
+      uiNotify(HOTMILK_SEEDED_MESSAGE, "info");
     }
 
     if (runtime.configError) {
-      ctx.ui.notify(
-        HOTMILK_PARSE_ERROR_MESSAGE(runtime.configPath, runtime.configError),
-        "warning",
-      );
+      uiNotify(HOTMILK_PARSE_ERROR_MESSAGE(runtime.configPath, runtime.configError), "warning");
     }
 
     if (runtime.extensionToggles["gentle-ai"]) {
       seedPersonaFromDefaults(ctx.cwd, runtime.defaults);
     }
 
-    const useContextModeExtension =
-      runtime.extensionToggles["context-mode"] && !runtime.extensionToggles["mcp-adapter"];
-    if (useContextModeExtension) {
-      const mcpPrune = pruneContextModeMcpServerFromAgentConfig();
-      if (mcpPrune.pruned) {
-        ctx.ui.notify(MCP_PRUNED_CONTEXT_MODE_MESSAGE(mcpPrune.path), "info");
-      }
+    applyContextStackOnSessionStart(runtime, uiNotify);
+
+    if (shouldWarnCavemanJaConflict(runtime.extensionToggles.caveman, runtime.defaults.language)) {
+      uiNotify(CAVEMAN_JA_CONFLICT_MESSAGE, "warning");
     }
 
     if (runtime.mcp.seedOnStart) {
       const mcpSeed = seedAgentMcpJsonIfMissing();
       if (mcpSeed.seeded) {
-        ctx.ui.notify(MCP_SEEDED_MESSAGE(mcpSeed.path), "info");
+        uiNotify(MCP_SEEDED_MESSAGE(mcpSeed.path), "info");
       }
     }
   });

package/src/config/extension-groups.ts

@@ -10,21 +10,16 @@ export const BUNDLED_EXTENSION_GROUPS: BundledExtensionGroup[] = [
   { label: "Harness", ids: ["skill-registry", "sdd-init", "gentle-ai"] },
   {
     label: "Agent tools",
-    ids: [
-      "context-mode",
-      "ask-user",
-      "graphify",
-      "subagents",
-      "agent-dashboard",
-      "web-access",
-      "pi-flows",
-    ],
+    ids: ["ask-user", "graphify", "subagents", "agent-dashboard", "web-access", "pi-flows"],
+  },
+  {
+    label: "Context & performance",
+    ids: ["context-mode", "rtk-optimizer", "simplify"],
   },
   {
     label: "Integrations",
     ids: ["goal", "docparser", "obsidian", "cursor-provider", "btw", "mcp-adapter"],
   },
-  { label: "Performance", ids: ["simplify", "rtk-optimizer"] },
   { label: "Workflow", ids: ["planning-with-files", "red-green"] },
   { label: "Output", ids: ["caveman"] },
 ];

package/src/index.ts

@@ -6,18 +6,25 @@ import type { ExtensionAPI } from "@earendil-works/pi-coding-agent";
 import { createHotmilkRuntime } from "./config/runtime.ts";
 import { registerDefaultsHandlers } from "./bootstrap/defaults.ts";
 import { registerBundledExtensions } from "./bootstrap/extensions.ts";
+import { prepareContextStack } from "./bootstrap/context-stack.ts";
 import { registerGraphHandlers } from "./bootstrap/graph.ts";
 import { registerSessionHandlers } from "./bootstrap/session.ts";
 import { registerInputCommands, routeInputCommand } from "./controller/input.ts";
+import { registerHotmilkSessionLogo } from "./ui/session-logo.ts";
 
 export default async function registerHotmilk(pi: ExtensionAPI): Promise<void> {
   const runtime = createHotmilkRuntime();
 
+  // Register before bundled imports so session_start handlers exist when bindExtensions emits.
+  registerHotmilkSessionLogo(pi);
+
   if (runtime.extensionToggles["agent-dashboard"]) {
     const { scheduleDashboardWarmStart } = await import("./bootstrap/dashboard.ts");
     scheduleDashboardWarmStart();
   }
 
+  prepareContextStack(runtime.extensionToggles);
+
   await registerBundledExtensions(pi, runtime.extensionToggles);
   registerGraphHandlers(pi, runtime.graph);
   registerDefaultsHandlers(pi, runtime.defaults);

package/src/ui/footer.ts

@@ -4,10 +4,51 @@ import {
   type ReadonlyFooterDataProvider,
 } from "@earendil-works/pi-coding-agent";
 import type { ThinkingLevel } from "@earendil-works/pi-agent-core";
-import { truncateToWidth } from "@earendil-works/pi-tui";
+import { truncateToWidth, visibleWidth } from "@earendil-works/pi-tui";
 import type { ExtensionContext } from "../controller/context.ts";
 import { formatFooterTime } from "./footer-time.ts";
 
+function sanitizeStatusText(text: string): string {
+  return text
+    .replace(/[\r\n\t]/g, " ")
+    .replace(/ +/g, " ")
+    .trim();
+}
+
+function extensionStatusLines(
+  footerData: ReadonlyFooterDataProvider,
+  width: number,
+  dim: (text: string) => string,
+  ellipsis: string,
+): string[] {
+  const extensionStatuses = footerData.getExtensionStatuses();
+  if (extensionStatuses.size === 0) {
+    return [];
+  }
+
+  return [...extensionStatuses.entries()]
+    .sort(([a], [b]) => a.localeCompare(b))
+    .map(([, text]) => sanitizeStatusText(text))
+    .filter((text) => text.length > 0)
+    .map((text) => truncateToWidth(dim(text), width, ellipsis));
+}
+
+function appendMetaToLastLine(lines: string[], meta: string, width: number): string[] {
+  if (lines.length === 0) {
+    return [meta];
+  }
+
+  const lastIndex = lines.length - 1;
+  const lastLine = lines[lastIndex] ?? "";
+  const gap = 2;
+  if (visibleWidth(lastLine) + gap + visibleWidth(meta) <= width) {
+    const padding = " ".repeat(width - visibleWidth(lastLine) - visibleWidth(meta));
+    return [...lines.slice(0, lastIndex), lastLine + padding + meta];
+  }
+
+  return [...lines, meta];
+}
+
 const FOOTER_TIME_REFRESH_MS = 30_000;
 
 function latestThinkingLevel(sessionManager: ExtensionContext["sessionManager"]): ThinkingLevel {
@@ -55,9 +96,17 @@ export function setupHotmilkFooter(ctx: ExtensionContext, termProgram: string):
         base.invalidate();
       },
       render(width: number): string[] {
-        const lines = base.render(width);
-        const meta = theme.fg("dim", `${formatFooterTime(new Date())}  ${termProgram}`);
-        return [...lines, "", truncateToWidth(meta, width, theme.fg("dim", "..."))];
+        const baseLines = base.render(width);
+        const coreLines = baseLines.slice(0, 2);
+        const dim = (text: string) => theme.fg("dim", text);
+        const ellipsis = theme.fg("dim", "...");
+        const statusLines = extensionStatusLines(footerData, width, dim, ellipsis);
+        const meta = truncateToWidth(
+          dim(`${formatFooterTime(new Date())}  ${termProgram}`),
+          width,
+          ellipsis,
+        );
+        return appendMetaToLastLine([...coreLines, ...statusLines], meta, width);
       },
     };
   });

package/src/ui/session-logo.ts

@@ -0,0 +1,73 @@
+import type { ExtensionAPI } from "@earendil-works/pi-coding-agent";
+import type { TUI } from "@earendil-works/pi-tui";
+import type { ExtensionContext } from "../controller/context.ts";
+import { HOTMILK_LOGO } from "./logo.ts";
+
+const LOGO_LINES = HOTMILK_LOGO.split("\n");
+/** Same lead time as gentle-pi `startup-banner.ts`. */
+const LOGO_SHOW_DELAY_MS = 50;
+
+let showTimeout: ReturnType<typeof setTimeout> | undefined;
+let activeTui: TUI | undefined;
+
+/** Center fixed-width ASCII lines within the terminal width. */
+export function centerAsciiLines(lines: readonly string[], width: number): string[] {
+  const contentWidth = Math.max(...lines.map((line) => line.length), 1);
+  const pad = Math.max(0, Math.floor((width - contentWidth) / 2));
+  const prefix = " ".repeat(pad);
+  return lines.map((line) => prefix + line.padEnd(contentWidth));
+}
+
+/** Skip only `/reload`; show on fresh launch and session resume (gentle-pi shows every session_start). */
+export function shouldShowSessionLogo(reason: string): boolean {
+  return reason !== "reload";
+}
+
+function clearLogoTimers(): void {
+  if (showTimeout) {
+    clearTimeout(showTimeout);
+    showTimeout = undefined;
+  }
+}
+
+function dismissHotmilkLogo(ctx: ExtensionContext): void {
+  if (!ctx.hasUI) return;
+  ctx.ui.setHeader(undefined);
+  activeTui?.requestRender();
+  activeTui = undefined;
+}
+
+function showHotmilkSessionLogo(ctx: ExtensionContext): void {
+  if (!ctx.hasUI) return;
+
+  clearLogoTimers();
+  showTimeout = setTimeout(() => {
+    showTimeout = undefined;
+    if (!ctx.hasUI) return;
+
+    ctx.ui.setHeader((tui, theme) => {
+      activeTui = tui;
+      return {
+        render(width: number): string[] {
+          return centerAsciiLines(LOGO_LINES, width).map((line) => theme.fg("accent", line));
+        },
+        invalidate() {
+          clearLogoTimers();
+        },
+      };
+    });
+  }, LOGO_SHOW_DELAY_MS);
+}
+
+/** Register persistent header logo on session start; clear on shutdown only. */
+export function registerHotmilkSessionLogo(pi: ExtensionAPI): void {
+  pi.on("session_start", (event, ctx) => {
+    if (!shouldShowSessionLogo(event.reason)) return;
+    showHotmilkSessionLogo(ctx);
+  });
+
+  pi.on("session_shutdown", (_event, ctx) => {
+    clearLogoTimers();
+    dismissHotmilkLogo(ctx);
+  });
+}