@sma1lboy/kobe 0.5.22 → 0.5.25

package/dist/share/skills/kobe/SKILL.md

@@ -0,0 +1,109 @@
+---
+name: kobe
+description: Spawn and coordinate parallel coding tasks via the kobe TUI from your shell. Use when the user asks to "try N approaches in parallel", "fan out", "compare implementations side-by-side", or "split this into subtasks". One task per attempt, each gets its own git worktree and its own agent session.
+---
+
+# kobe — parallel coding tasks from your shell
+
+kobe is a local terminal UI that runs many AI coding sessions at once.
+Each task lives in its own git worktree with its own agent session.
+When the user asks for parallel exploration, fan out by spawning kobe
+tasks instead of doing N attempts sequentially in one chat.
+
+The daemon must already be running. The TUI auto-starts it; if not, run
+`kobe daemon start` once.
+
+## Triggers — fire fan-out when the user says
+
+- "try N approaches in parallel", "并行试 N 个"
+- "fan out", "split this into subtasks"
+- "compare X and Y side-by-side"
+- "explore a few different ways to..."
+- explicit count: "spin up 3 tasks for..."
+
+Single ambiguous tasks → no fan-out. Just do them in your own chat.
+
+## How — the five verbs
+
+Each verb is a one-shot shell command. Reads stdin: none. Writes a JSON
+object to stdout, exits 0. On error, writes
+`{"error":{"message":"...","code":"..."}}` to stderr and exits non-zero.
+
+```bash
+# Spawn a new task. Returns task JSON; read `.taskId`.
+kobe api spawn-task --repo $PWD --prompt "<scoped prompt>" [--title T] [--base-branch B]
+
+# Create an extra chat tab on an existing task.
+kobe api create-tab --task-id <id> [--title T]
+
+# Send a follow-up prompt to a task (resumes the agent session).
+kobe api send --task-id <id> --prompt "<text>" [--tab-id TID]
+
+# Read a task's current state (status, branch, worktree, tabs).
+kobe api get-task --task-id <id>
+
+# Read a single tab off a task.
+kobe api get-tab --task-id <id> --tab-id <tab-id>
+```
+
+Add `--pretty` to any verb to pretty-print stdout for inspection.
+Output to stdout is always one JSON object terminated with `\n`.
+
+## Workflow — fan-out + report back
+
+```bash
+# 1. Spawn 3 parallel attempts. Cap at 3-4.
+T1=$(kobe api spawn-task --repo "$PWD" --prompt "Approach A: use a state machine" | jq -r .taskId)
+T2=$(kobe api spawn-task --repo "$PWD" --prompt "Approach B: use event sourcing"    | jq -r .taskId)
+T3=$(kobe api spawn-task --repo "$PWD" --prompt "Approach C: use a reducer pattern" | jq -r .taskId)
+
+# 2. Tell the user what you spawned so the sidebar reads back correctly.
+echo "Spawned three tasks: $T1 (state machine), $T2 (event sourcing), $T3 (reducer)"
+
+# 3. Poll until each is idle. Don't tight-loop; sleep 5-15s between checks.
+for ID in $T1 $T2 $T3; do
+  while true; do
+    STATUS=$(kobe api get-task --task-id "$ID" | jq -r .task.status)
+    case "$STATUS" in
+      idle|awaiting-approval|done) break ;;
+    esac
+    sleep 10
+  done
+done
+
+# 4. Read each result and aggregate. The active tab's id is on the task.
+for ID in $T1 $T2 $T3; do
+  TAB=$(kobe api get-task --task-id "$ID" | jq -r .task.activeTabId)
+  kobe api get-tab --task-id "$ID" --tab-id "$TAB"
+done
+```
+
+## Do
+
+- Cap fan-out at 3-4 in parallel. The orchestrator caps total concurrency at 4.
+- Give each subtask its own scoped prompt — don't dump the whole parent conversation.
+- Tell the user what was spawned, with IDs, so they can follow along in the sidebar.
+- Poll status with `sleep` between calls. 5-15 seconds is fine.
+- Aggregate results back into your own chat before reporting to the user.
+
+## Don't
+
+- Don't fan out for single simple tasks. One thing → do it yourself in chat.
+- Don't recursively spawn from inside a spawned task. There is no recursion
+  guard yet; you will starve the concurrency cap and the inner spawn will
+  hang.
+- Don't use `kobe api send` as a chat channel. Every send is a full agent
+  turn — it costs tokens and time. Send a complete instruction, not a
+  conversation.
+- Don't delete tasks. There is no `kobe api delete` verb on purpose. Cleanup
+  happens in the TUI by the user.
+
+## When the daemon is not running
+
+`kobe api ...` exits 2 with `{"error":{"code":"BAD_DAEMON",...}}` on stderr.
+If you see this:
+
+> The kobe daemon is not running. Ask the user to run `kobe daemon start`
+> (or launch the kobe TUI, which auto-starts it), then re-try.
+
+Don't try to start the daemon yourself; that's the user's call.

package/package.json

@@ -1,13 +1,12 @@
 {
   "$schema": "https://json.schemastore.org/package.json",
   "name": "@sma1lboy/kobe",
-  "version": "0.5.22",
+  "version": "0.5.25",
   "description": "TUI orchestrator for Claude Code (codename)",
   "type": "module",
   "packageManager": "bun@1.3.13",
   "bin": {
-    "kobe": "dist/cli/index.js",
-    "kobed": "dist/bin/kobed.js"
+    "kobe": "dist/cli/index.js"
   },
   "files": ["dist", "README.md", "LICENSE"],
   "publishConfig": {
@@ -38,6 +37,7 @@
     "test:socket": "KOBE_INCLUDE_SOCKET=1 vitest run test/daemon test/orchestrator/bridge.test.ts --pool forks --minWorkers=1 --maxWorkers=1 --passWithNoTests",
     "test:behavior": "KOBE_INCLUDE_BEHAVIOR=1 vitest run test/behavior --pool forks --minWorkers=1 --maxWorkers=1 --passWithNoTests",
     "lint": "biome check .",
+    "postinstall": "bun run scripts/check-preview-deps.ts || true",
     "prepublishOnly": "bun run typecheck && bun run build"
   },
   "//": "biome.json + bunfig.toml live at the monorepo root since they apply repo-wide; bun.lock also lives at root (workspace-shared).",