create extension and skill to run yaks

.pi/extensions/yak-mode-gate.ts

@@ -0,0 +1,19 @@
+import type { ExtensionAPI } from "@earendil-works/pi-coding-agent";
+import { isToolCallEventType } from "@earendil-works/pi-coding-agent";
+
+export default function (pi: ExtensionAPI) {
+  pi.on("tool_call", async (event, ctx) => {
+    if (!isToolCallEventType("bash", event)) return;
+
+    const cmd = event.input.command || "";
+    const isYxCommand = cmd.includes("yx ");
+
+    // Allow yaks in interactive mode, block in print mode (sub-agents)
+    if (isYxCommand && ctx.mode === "print") {
+      return {
+        block: true,
+        reason: "yx commands are disabled in print mode. Sub-agents must focus on domain work only.",
+      };
+    }
+  });
+}

.pi/skills/yak-tasks/SKILL.md

@@ -0,0 +1,84 @@
+---
+name: yak-tasks
+description: Orchestrate sub-agents using yx task tracking. Use when delegating work to sub-agents, tracking task progress, or marking tasks complete. For breaking down problems into tasks, use /skill:problem-breakdown first.
+---
+
+# Yak Task Orchestration
+
+Coordinate sub-agent work using the `yx` CLI. The main agent creates tasks and delegates to sub-agents; sub-agents execute domain work without yx CLI access.
+
+## Workflow
+
+1. **Break down** → `/skill:problem-breakdown` (creates yaks with context)
+2. **Delegate** → sub-agents run on yaks via `pi -p` (blocked from yx CLI)
+3. **Track** → `yx list` / `yx show <task-id>`
+4. **Verify** → run `npm run checks` after sub-agent work
+5. **Complete** → `yx done <task-id>`
+
+## Running Sub-Agents
+
+Sub-agents execute in print mode and are **hard-blocked from using `yx` commands** (see `.pi/extensions/yak-mode-gate.ts`). They receive:
+- The yak's `.context.md` (task description)
+- `AGENTS.md` (project conventions)
+- Domain files to work on
+- No yx CLI access
+
+```bash
+# Run a sub-agent on a specific yak
+# The sub-agent reads the yak context and executes the work
+pi -p "Work on yak: <task-name>. Read .yaks/<task-id>/.context.md for details."
+```
+
+## Tracking Progress
+
+```bash
+# List all yaks with hierarchy
+yx list
+
+# Show yak details
+yx show <task-id>
+
+# Check state directly
+cat .yaks/<task-id>/.state
+```
+
+States:
+- `pending` — not yet started
+- `in-progress` — being worked on
+- `done` — completed
+
+## Marking Tasks Complete
+
+```bash
+yx done <task-id>
+```
+
+Always verify before marking done:
+```bash
+# 1. Check state
+cat .yaks/<task-id>/.state
+
+# 2. Review what was done
+cat .yaks/<task-id>/.context.md
+
+# 3. Run project checks
+npm run checks
+```
+
+## Sub-Agent Communication
+
+Sub-agents can read yak state files directly (no yx CLI needed):
+```bash
+cat .yaks/<task-id>/.name    # task name
+cat .yaks/<task-id>/.state   # current state
+cat .yaks/<task-id>/.context.md  # task description
+cat .yaks/<task-id>/.created.json  # creation metadata
+```
+
+## Best Practices
+
+- **Delegate after breakdown** — run `/skill:problem-breakdown` first to create structured yaks
+- **Verify before marking done** — always run `npm run checks` to catch sub-agent errors
+- **Review context** — read `.context.md` to understand what the sub-agent was supposed to do
+- **Keep yaks focused** — each yak should be a single file operation or one method implementation
+- **Use hierarchy** — parent yaks block children; fix leaves first