diff --git a/extensions/caveman/index.ts b/extensions/caveman/index.ts
new file mode 100644
index 0000000..1c4bb7d
--- /dev/null
+++ b/extensions/caveman/index.ts
@@ -0,0 +1,226 @@
+/**
+ * pi-caveman — Compressed-output mode for Pi coding agent.
+ *
+ * Registers a `/caveman` slash command that toggles caveman-speak
+ * compression levels and persists state across session restarts.
+ *
+ * Usage:
+ *   /caveman             — toggle between off and full
+ *   /caveman lite|full|ultra|wenyan-lite|wenyan-full|wenyan-ultra|off|status
+ */
+
+import type { ExtensionAPI } from "@mariozechner/pi-coding-agent";
+import * as fs from "node:fs";
+import * as path from "node:path";
+
+// ── Types ────────────────────────────────────────────────────────────────
+
+type CavemanLevel = "lite" | "full" | "ultra" | "wenyan-lite" | "wenyan-full" | "wenyan-ultra";
+
+interface CavemanState {
+  active: boolean;
+  level: CavemanLevel;
+}
+
+// ── Constants ────────────────────────────────────────────────────────────
+
+const DEFAULT_LEVEL: CavemanLevel = "full";
+
+const STATE_DIR = path.join(
+  process.env.XDG_STATE_HOME ?? path.join(process.env.HOME!, ".local", "state"),
+  "pi-caveman",
+);
+
+const STATE_FILE = path.join(STATE_DIR, "caveman-state.json");
+
+const CAVEMAN_SKILL_PATH = path.join(
+  process.env.HOME!,
+  "ai-assets",
+  "skills",
+  "caveman",
+  "SKILL.md",
+);
+
+// ── State helpers ────────────────────────────────────────────────────────
+
+function readState(): CavemanState {
+  try {
+    const raw = fs.readFileSync(STATE_FILE, "utf-8");
+    const parsed = JSON.parse(raw);
+    return {
+      active: typeof parsed.active === "boolean" ? parsed.active : false,
+      level: isValidLevel(parsed.level) ? parsed.level : DEFAULT_LEVEL,
+    };
+  } catch {
+    return { active: false, level: DEFAULT_LEVEL };
+  }
+}
+
+function writeState(state: CavemanState): void {
+  fs.mkdirSync(STATE_DIR, { recursive: true });
+  fs.writeFileSync(STATE_FILE, JSON.stringify(state, null, 2), "utf-8");
+}
+
+function isValidLevel(v: unknown): v is CavemanLevel {
+  return (
+    typeof v === "string" &&
+    ["lite", "full", "ultra", "wenyan-lite", "wenyan-full", "wenyan-ultra"].includes(v)
+  );
+}
+
+// ── Skill content cache ──────────────────────────────────────────────────
+
+let cachedSkillContent: string | null = null;
+
+function getSkillContent(): string {
+  if (cachedSkillContent !== null) return cachedSkillContent;
+  try {
+    cachedSkillContent = fs.readFileSync(CAVEMAN_SKILL_PATH, "utf-8");
+    return cachedSkillContent;
+  } catch {
+    return "";
+  }
+}
+
+// ── Extension ────────────────────────────────────────────────────────────
+
+export default function (pi: ExtensionAPI): void {
+  // ── State ──
+  let state = readState();
+
+  function applyMode(): void {
+    writeState(state);
+    if (state.active) {
+      pi.setStatus("caveman", `🦴 caveman:${state.level}`);
+      // Inject caveman instructions into the system prompt each turn
+      pi.on("before_agent_start", async (_event) => {
+        if (!state.active) return;
+        const skill = getSkillContent();
+        if (!skill) return;
+        return {
+          systemPrompt: `${_event.systemPrompt}\n\n---\nCRITICAL OUTPUT MODE: You are in CAVEMAN MODE. Follow these rules:\n\n${skill}`,
+        };
+      });
+    } else {
+      // Clear any previous status
+      // (constructor-style state means we can't easily unregister)
+    }
+  }
+
+  // ── Footer badge (always shown) ──
+  function updateFooter(): void {
+    if (state.active) {
+      pi.setStatus("caveman", `🦴 caveman:${state.level}`);
+    } else {
+      pi.setStatus("caveman", "");
+    }
+  }
+
+  // ── /caveman command ────────────────────────────────────────────────
+
+  pi.registerCommand("caveman", {
+    description: "Toggle caveman compressed-output mode (lite|full|ultra|wenyan-lite|wenyan-full|wenyan-ultra|off|status)",
+    handler: async (args, ctx) => {
+      const arg = args?.trim().toLowerCase() ?? "";
+
+      switch (arg) {
+        case "":
+          // Toggle: off → full, anything → off
+          if (state.active) {
+            state.active = false;
+            ctx.ui.notify("🦴 Caveman: off", "info");
+          } else {
+            state.active = true;
+            state.level = DEFAULT_LEVEL;
+            ctx.ui.notify(`🦴 Caveman: ${state.level}`, "info");
+          }
+          break;
+
+        case "off":
+        case "disable":
+          state.active = false;
+          ctx.ui.notify("🦴 Caveman: off", "info");
+          break;
+
+        case "status":
+          ctx.ui.notify(
+            state.active ? `🦴 Caveman: ${state.level} (active)` : "🦴 Caveman: off",
+            "info",
+          );
+          return; // Don't update footer (already correct)
+
+        case "lite":
+        case "full":
+        case "ultra":
+        case "wenyan-lite":
+        case "wenyan-full":
+        case "wenyan-ultra":
+          state.active = true;
+          state.level = arg as CavemanLevel;
+          ctx.ui.notify(`🦴 Caveman: ${state.level}`, "info");
+          break;
+
+        default:
+          ctx.ui.notify(
+            `🦴 Unknown level: "${arg}". Use: lite, full, ultra, wenyan-lite, wenyan-full, wenyan-ultra, off, status`,
+            "error",
+          );
+          return;
+      }
+
+      writeState(state);
+      updateFooter();
+
+      // Inject skill content as a user message when activating
+      if (state.active) {
+        const skill = getSkillContent();
+        if (skill) {
+          pi.sendUserMessage(
+            `/skill:caveman`,
+            { deliverAs: "nextTurn" },
+          );
+        }
+      }
+    },
+  });
+
+  // ── Startup ─────────────────────────────────────────────────────────
+
+  pi.on("session_start", async () => {
+    // Re-read state (may have changed externally)
+    state = readState();
+    updateFooter();
+
+    // If active, inject skill into agent context
+    if (state.active) {
+      const skill = getSkillContent();
+      if (skill) {
+        pi.appendEntry("caveman-active", { level: state.level });
+      }
+    }
+  });
+
+  // ── Inject caveman rules when active ────────────────────────────────
+
+  pi.on("before_agent_start", async (_event) => {
+    // Re-read state in case it changed
+    const current = readState();
+    if (!current.active) return;
+
+    const skill = getSkillContent();
+    if (!skill) return;
+
+    return {
+      systemPrompt: `${_event.systemPrompt}\n\n---\nCRITICAL OUTPUT MODE: You are in CAVEMAN MODE at level "${current.level}". Follow these rules exactly:\n\n${skill}`,
+    };
+  });
+
+  // ── Cleanup on shutdown ─────────────────────────────────────────────
+
+  pi.on("session_shutdown", async () => {
+    writeState(state);
+  });
+
+  // ── Apply initial mode ──────────────────────────────────────────────
+  applyMode();
+}
diff --git a/extensions/caveman/package.json b/extensions/caveman/package.json
new file mode 100644
index 0000000..51508f6
--- /dev/null
+++ b/extensions/caveman/package.json
@@ -0,0 +1,16 @@
+{
+  "name": "pi-caveman",
+  "version": "1.0.0",
+  "description": "Caveman compressed-output mode for Pi coding agent — toggle with /caveman",
+  "license": "MIT",
+  "type": "module",
+  "main": "./index.ts",
+  "pi": {
+    "extensions": [
+      "./index.ts"
+    ]
+  },
+  "peerDependencies": {
+    "@mariozechner/pi-coding-agent": "*"
+  }
+}
diff --git a/skills/caveman/SKILL.md b/skills/caveman/SKILL.md
new file mode 100644
index 0000000..835dd0d
--- /dev/null
+++ b/skills/caveman/SKILL.md
@@ -0,0 +1,79 @@
+---
+name: caveman
+description: >
+  Ultra-compressed communication mode. Cuts token usage ~75% by speaking like caveman
+  while keeping full technical accuracy. Supports intensity levels: lite, full (default), ultra,
+  wenyan-lite, wenyan-full, wenyan-ultra.
+  Use when user says "caveman mode", "talk like caveman", "use caveman", "less tokens",
+  "be brief", or invokes /caveman. Also auto-triggers when token efficiency is requested.
+---
+
+Terse like smart caveman. All technical substance stay. Only fluff die.
+
+## Persistence
+
+ACTIVE EVERY RESPONSE. No revert after many turns. No filler drift. Still active if unsure.
+Off only: "stop caveman" / "normal mode".
+
+Default: **full**. Switch: `/caveman lite|full|ultra`.
+
+## Rules
+
+Drop: articles (a/an/the), filler (just/really/basically/actually/simply), pleasantries
+(sure/certainly/of course/happy to), hedging. Fragments OK. Short synonyms (big not extensive,
+fix not "implement a solution for"). Technical terms exact. Code blocks unchanged.
+Errors quoted exact.
+
+Pattern: `[thing] [action] [reason]. [next step].`
+
+Not: "Sure! I'd be happy to help you with that. The issue you're experiencing is likely caused by..."
+Yes: "Bug in auth middleware. Token expiry check use `<` not `<=`. Fix:"
+
+## Intensity
+
+| Level | What change |
+|-------|------------|
+| **lite** | No filler/hedging. Keep articles + full sentences. Professional but tight |
+| **full** | Drop articles, fragments OK, short synonyms. Classic caveman |
+| **ultra** | Abbreviate prose words (DB/auth/config/req/res/fn/impl), strip conjunctions, arrows for causality (X → Y), one word when one word enough. Code symbols, function names, API names, error strings: never abbreviate |
+| **wenyan-lite** | Semi-classical. Drop filler/hedging but keep grammar structure, classical register |
+| **wenyan-full** | Maximum classical terseness. Fully 文言文. 80-90% character reduction. Classical sentence patterns, verbs precede objects, subjects often omitted, classical particles (之/乃/為/其) |
+| **wenyan-ultra** | Extreme abbreviation while keeping classical Chinese feel. Maximum compression, ultra terse |
+
+Example — "Why React component re-render?"
+- lite: "Your component re-renders because you create a new object reference each render. Wrap it in `useMemo`."
+- full: "New object ref each render. Inline object prop = new ref = re-render. Wrap in `useMemo`."
+- ultra: "Inline obj prop → new ref → re-render. `useMemo`."
+- wenyan-lite: "組件頻重繪，以每繪新生對象參照故。以 useMemo 包之。"
+- wenyan-full: "物出新參照，致重繪。useMemo Wrap之。"
+- wenyan-ultra: "新參照→重繪。useMemo Wrap。"
+
+Example — "Explain database connection pooling."
+- lite: "Connection pooling reuses open connections instead of creating new ones per request. Avoids repeated handshake overhead."
+- full: "Pool reuse open DB connections. No new connection per request. Skip handshake overhead."
+- ultra: "Pool = reuse DB conn. Skip handshake → fast under load."
+- wenyan-full: "池reuse open connection。不每req新開。skip handshake overhead。"
+- wenyan-ultra: "池reuse conn。skip handshake → fast。"
+
+## Auto-Clarity
+
+Drop caveman when:
+- Security warnings
+- Irreversible action confirmations
+- Multi-step sequences where fragment order or omitted conjunctions risk misread
+- Compression itself creates technical ambiguity (e.g., "migrate table drop column backup first" — order unclear without articles/conjunctions)
+- User asks to clarify or repeats question
+
+Resume caveman after clear part done.
+
+Example — destructive op:
+> **Warning:** This will permanently delete all rows in the `users` table and cannot be undone.
+> ```sql
+> DROP TABLE users;
+> ```
+> Caveman resume. Verify backup exist first.
+
+## Boundaries
+
+Code/commits/PRs: write normal. "stop caveman" or "normal mode": revert.
+Level persist until changed or session end.