prizmkit 1.0.26 → 1.0.28

package/bundled/VERSION.json

@@ -1,5 +1,5 @@
 {
-  "frameworkVersion": "1.0.26",
-  "bundledAt": "2026-03-15T17:57:13.772Z",
-  "bundledFrom": "896326f"
+  "frameworkVersion": "1.0.28",
+  "bundledAt": "2026-03-17T06:14:03.182Z",
+  "bundledFrom": "780f717"
 }

package/bundled/dev-pipeline/README.md

@@ -33,6 +33,7 @@ python3 dev-pipeline/scripts/init-pipeline.py \
 |---------|-------------|
 | `./run.sh run [feature-list.json] [options]` | Start or resume the pipeline. Processes features sequentially by dependency order. |
 | `./run.sh status [feature-list.json]` | Display current pipeline status: completed, pending, blocked, failed features. |
+| `./run.sh test-cli` | Test AI CLI detection: show detected CLI, version, platform, and query the AI model identity. |
 | `./run.sh reset` | Clear all runtime state in `state/`. Pipeline starts fresh on next `run`. |
 | `./run.sh help` | Show usage help. |
 | `./retry-feature.sh <feature-id> [feature-list.json]` | Retry a single failed feature. Runs one session then exits. |
@@ -97,7 +98,9 @@ What is always reset (with or without `--clean`):
 | `MAX_RETRIES` | `3` | Maximum retry attempts per feature before marking as failed. |
 | `SESSION_TIMEOUT` | `0` (no limit) | Timeout in seconds per AI CLI session. 0 = no timeout. |
 | `AI_CLI` | auto-detect | AI CLI command name. Auto-detects `cbc` or `claude`. Set to override. |
+| `MODEL` | (none) | AI model ID for the session. Passed as `--model` to the CLI. See [Model Selection](#model-selection). |
 | `CODEBUDDY_CLI` | (deprecated) | Legacy alias for `AI_CLI`. Prefer `AI_CLI`. |
+| `VERBOSE` | `0` | Set to `1` to enable `--verbose` on AI CLI (shows subagent output). |
 | `HEARTBEAT_INTERVAL` | `30` | Seconds between heartbeat log output while a session is running. |
 | `HEARTBEAT_STALE_THRESHOLD` | `600` | Seconds before a session is considered stale/stuck. |
 | `LOG_CLEANUP_ENABLED` | `1` | Run log cleanup before pipeline execution (`1`=enabled, `0`=disabled). |
@@ -116,6 +119,93 @@ SESSION_TIMEOUT=7200 ./dev-pipeline/run.sh run feature-list.json
 LOG_RETENTION_DAYS=7 LOG_MAX_TOTAL_MB=512 ./dev-pipeline/run.sh run feature-list.json
 ```
 
+### AI CLI Configuration
+
+The pipeline auto-detects which AI CLI to use. Detection priority:
+
+1. `AI_CLI` environment variable (highest)
+2. `.prizmkit/config.json` → `ai_cli` field
+3. `CODEBUDDY_CLI` environment variable (legacy)
+4. Auto-detect: `cbc` in PATH → `claude` in PATH (lowest)
+
+To permanently configure a project to use a specific CLI, create `.prizmkit/config.json`:
+
+```json
+{
+  "ai_cli": "claude-internal",
+  "platform": "claude"
+}
+```
+
+Or override per-invocation:
+
+```bash
+AI_CLI=claude-internal ./dev-pipeline/run.sh run feature-list.json
+```
+
+### Model Selection
+
+Use the `MODEL` environment variable to specify which AI model to use. The value is passed as `--model <id>` to the CLI.
+
+```bash
+# Run pipeline with Sonnet (faster, cheaper)
+MODEL=claude-sonnet-4.6 ./dev-pipeline/run.sh run feature-list.json
+
+# Run pipeline with Opus (most capable)
+MODEL=claude-opus-4.6 ./dev-pipeline/run.sh run feature-list.json
+
+# Retry a feature with a specific model
+MODEL=claude-opus-4.6 ./dev-pipeline/retry-feature.sh F-007
+
+# Test which model the CLI is using
+MODEL=claude-sonnet-4.6 ./dev-pipeline/run.sh test-cli
+```
+
+Common model IDs (for `cbc`):
+
+| Model ID | Description |
+|----------|-------------|
+| `claude-opus-4.6` | Most capable, slower, higher cost |
+| `claude-sonnet-4.6` | Balanced speed/capability (recommended for pipeline) |
+| `claude-haiku-4.5` | Fastest, cheapest, less capable |
+
+> **Note**: `--model` support depends on the CLI. `cbc` fully supports it. `claude-internal` does not support `--model` in headless mode (only interactive `/model` command). If `MODEL` is set but the CLI doesn't support it, the flag is silently ignored.
+
+### Testing AI CLI (`test-cli`)
+
+Use `test-cli` to verify which CLI, version, and model the pipeline will use:
+
+```bash
+# Basic test — uses auto-detected CLI and default model
+./dev-pipeline/run.sh test-cli
+
+# Test with a specific model
+MODEL=claude-sonnet-4.6 ./dev-pipeline/run.sh test-cli
+
+# Test with a specific CLI
+AI_CLI=cbc ./dev-pipeline/run.sh test-cli
+```
+
+Example output:
+
+```
+============================================
+  Dev-Pipeline AI CLI Test
+============================================
+
+  Detected CLI:    cbc
+  Platform:        codebuddy
+  CLI Version:     2.62.1
+
+  Querying AI model (headless mode)...
+
+  AI Response:     I'm CodeBuddy, running Claude Opus 4.6
+
+============================================
+```
+
+The test sends a one-line prompt asking the AI to identify itself, with a 30-second timeout. If the CLI requires authentication or is unavailable, it shows a fallback message.
+
 ## How It Works
 
 ### Execution Flow

package/bundled/dev-pipeline/retry-bug.sh

@@ -227,6 +227,11 @@ if [[ "$USE_STREAM_JSON" == "true" ]]; then
 fi
 
 # Spawn AI CLI session
+MODEL_FLAG=""
+if [[ -n "${MODEL:-}" ]]; then
+    MODEL_FLAG="--model $MODEL"
+fi
+
 case "$CLI_CMD" in
     *claude*)
         "$CLI_CMD" \
@@ -234,6 +239,7 @@ case "$CLI_CMD" in
             -p "$(cat "$BOOTSTRAP_PROMPT")" \
             --yes \
             $STREAM_JSON_FLAG \
+            $MODEL_FLAG \
             > "$SESSION_LOG" 2>&1 &
         ;;
     *)
@@ -241,6 +247,7 @@ case "$CLI_CMD" in
             --print \
             -y \
             $STREAM_JSON_FLAG \
+            $MODEL_FLAG \
             < "$BOOTSTRAP_PROMPT" \
             > "$SESSION_LOG" 2>&1 &
         ;;

package/bundled/dev-pipeline/retry-feature.sh

@@ -228,6 +228,11 @@ python3 "$SCRIPTS_DIR/update-feature-status.py" \
     --action start >/dev/null 2>&1 || true
 
 # Spawn AI CLI session
+MODEL_FLAG=""
+if [[ -n "${MODEL:-}" ]]; then
+    MODEL_FLAG="--model $MODEL"
+fi
+
 case "$CLI_CMD" in
     *claude*)
         "$CLI_CMD" \
@@ -235,6 +240,7 @@ case "$CLI_CMD" in
             -p "$(cat "$BOOTSTRAP_PROMPT")" \
             --yes \
             $STREAM_JSON_FLAG \
+            $MODEL_FLAG \
             > "$SESSION_LOG" 2>&1 &
         ;;
     *)
@@ -242,6 +248,7 @@ case "$CLI_CMD" in
             --print \
             -y \
             $STREAM_JSON_FLAG \
+            $MODEL_FLAG \
             < "$BOOTSTRAP_PROMPT" \
             > "$SESSION_LOG" 2>&1 &
         ;;

package/bundled/dev-pipeline/run-bugfix.sh

@@ -79,6 +79,11 @@ spawn_and_wait_session() {
         stream_json_flag="--output-format stream-json"
     fi
 
+    local model_flag=""
+    if [[ -n "${MODEL:-}" ]]; then
+        model_flag="--model $MODEL"
+    fi
+
     case "$CLI_CMD" in
         *claude*)
             "$CLI_CMD" \
@@ -87,6 +92,7 @@ spawn_and_wait_session() {
                 --yes \
                 $verbose_flag \
                 $stream_json_flag \
+                $model_flag \
                 > "$session_log" 2>&1 &
             ;;
         *)
@@ -95,6 +101,7 @@ spawn_and_wait_session() {
                 -y \
                 $verbose_flag \
                 $stream_json_flag \
+                $model_flag \
                 < "$bootstrap_prompt" \
                 > "$session_log" 2>&1 &
             ;;

package/bundled/dev-pipeline/run.sh

@@ -20,6 +20,7 @@ set -euo pipefail
 #   AI_CLI                AI CLI command name (override; also readable from .prizmkit/config.json)
 #   CODEBUDDY_CLI         Legacy alias for AI_CLI (deprecated, use AI_CLI instead)
 #   PRIZMKIT_PLATFORM     Force platform: 'codebuddy' or 'claude' (auto-detected)
+#   MODEL                 AI model to use (e.g. claude-opus-4.6, claude-sonnet-4.6, claude-haiku-4.5)
 #   VERBOSE               Set to 1 to enable --verbose on AI CLI (shows subagent output)
 #   HEARTBEAT_INTERVAL    Heartbeat log interval in seconds (default: 30)
 #   HEARTBEAT_STALE_THRESHOLD  Heartbeat stale threshold in seconds (default: 600)
@@ -41,6 +42,7 @@ LOG_CLEANUP_ENABLED=${LOG_CLEANUP_ENABLED:-1}
 LOG_RETENTION_DAYS=${LOG_RETENTION_DAYS:-14}
 LOG_MAX_TOTAL_MB=${LOG_MAX_TOTAL_MB:-1024}
 VERBOSE=${VERBOSE:-0}
+MODEL=${MODEL:-""}
 
 # Source shared common helpers (CLI/platform detection + logs + deps)
 source "$SCRIPT_DIR/lib/common.sh"
@@ -91,6 +93,11 @@ spawn_and_wait_session() {
         stream_json_flag="--output-format stream-json"
     fi
 
+    local model_flag=""
+    if [[ -n "$MODEL" ]]; then
+        model_flag="--model $MODEL"
+    fi
+
     case "$CLI_CMD" in
         *claude*)
             # Claude Code: prompt via -p argument, --yes for auto-accept
@@ -100,6 +107,7 @@ spawn_and_wait_session() {
                 --yes \
                 $verbose_flag \
                 $stream_json_flag \
+                $model_flag \
                 > "$session_log" 2>&1 &
             ;;
         *)
@@ -109,6 +117,7 @@ spawn_and_wait_session() {
                 -y \
                 $verbose_flag \
                 $stream_json_flag \
+                $model_flag \
                 < "$bootstrap_prompt" \
                 > "$session_log" 2>&1 &
             ;;
@@ -790,6 +799,7 @@ show_help() {
     echo "  run [feature-list.json]                 Run all features sequentially"
     echo "  run <feature-id> [options]              Run a single feature"
     echo "  status [feature-list.json]               Show pipeline status"
+    echo "  test-cli                                 Test AI CLI: show detected CLI, version, and model"
     echo "  reset                                    Clear all state and start fresh"
     echo "  help                                     Show this help message"
     echo ""
@@ -805,6 +815,7 @@ show_help() {
     echo "  MAX_RETRIES           Max retries per feature (default: 3)"
     echo "  SESSION_TIMEOUT       Session timeout in seconds (default: 0 = no limit)"
     echo "  AI_CLI                AI CLI command name (auto-detected: cbc or claude)"
+    echo "  MODEL                 AI model ID (e.g. claude-opus-4.6, claude-sonnet-4.6, claude-haiku-4.5)"
     echo "  HEARTBEAT_INTERVAL    Heartbeat log interval in seconds (default: 30)"
     echo "  HEARTBEAT_STALE_THRESHOLD  Heartbeat stale threshold in seconds (default: 600)"
     echo "  LOG_CLEANUP_ENABLED   Run log cleanup before execution (default: 1)"
@@ -820,6 +831,8 @@ show_help() {
     echo "  ./run.sh run F-007 --clean --mode standard             # Clean + run standard"
     echo "  ./run.sh status                                        # Show pipeline status"
     echo "  MAX_RETRIES=5 SESSION_TIMEOUT=7200 ./run.sh run        # Custom config"
+    echo "  MODEL=claude-sonnet-4.6 ./run.sh run                    # Use Sonnet model"
+    echo "  MODEL=claude-haiku-4.5 ./run.sh test-cli                # Test with Haiku"
 }
 
 case "${1:-run}" in
@@ -843,6 +856,64 @@ case "${1:-run}" in
             --state-dir "$STATE_DIR" \
             --action status
         ;;
+    test-cli)
+        echo ""
+        echo "============================================"
+        echo "  Dev-Pipeline AI CLI Test"
+        echo "============================================"
+        echo ""
+        echo "  Detected CLI:    $CLI_CMD"
+        echo "  Platform:        $PLATFORM"
+        if [[ -n "$MODEL" ]]; then
+            echo "  Requested Model: $MODEL"
+        fi
+
+        # Get CLI version (first line only)
+        cli_version=$("$CLI_CMD" -v 2>&1 | head -1 || echo "unknown")
+        echo "  CLI Version:     $cli_version"
+        echo ""
+        echo "  Querying AI model (headless mode)..."
+
+        test_prompt="What AI assistant/platform are you and what model are you running? Reply in one line, e.g. \"I'm Claude Code Claude Opnus x.x\".No extra text."
+
+        local_model_flag=""
+        if [[ -n "$MODEL" ]]; then
+            local_model_flag="--model $MODEL"
+        fi
+
+        # Run headless query with 30s timeout (background + kill pattern for macOS)
+        tmpfile=$(mktemp)
+        (
+            unset CLAUDECODE
+            case "$CLI_CMD" in
+                *claude*)
+                    "$CLI_CMD" --print -p "$test_prompt" --dangerously-skip-permissions --no-session-persistence $local_model_flag > "$tmpfile" 2>/dev/null
+                    ;;
+                *)
+                    echo "$test_prompt" | "$CLI_CMD" --print -y $local_model_flag > "$tmpfile" 2>/dev/null
+                    ;;
+            esac
+        ) &
+        query_pid=$!
+        ( sleep 30 && kill "$query_pid" 2>/dev/null ) &
+        timer_pid=$!
+        wait "$query_pid" 2>/dev/null
+        kill "$timer_pid" 2>/dev/null
+        wait "$timer_pid" 2>/dev/null || true
+
+        model_reply=$(cat "$tmpfile" 2>/dev/null | head -3)
+        rm -f "$tmpfile"
+
+        if [[ -z "$model_reply" ]]; then
+            model_reply="(no response — CLI may require auth or is unavailable)"
+        fi
+
+        echo ""
+        echo "  AI Response:     $model_reply"
+        echo ""
+        echo "============================================"
+        echo ""
+        ;;
     reset)
         log_warn "Resetting pipeline state..."
         rm -rf "$STATE_DIR"

package/bundled/dev-pipeline/scripts/init-pipeline.py

@@ -19,6 +19,7 @@ from datetime import datetime, timezone
 
 EXPECTED_SCHEMA = "dev-pipeline-feature-list-v1"
 FEATURE_ID_PATTERN = re.compile(r"^F-\d{3}$")
+TERMINAL_STATUSES = {"completed", "failed", "skipped"}
 
 REQUIRED_FEATURE_FIELDS = [
     "id",
@@ -234,6 +235,12 @@ def create_state_directory(state_dir, feature_list_path, features):
     os.makedirs(abs_state_dir, exist_ok=True)
     os.makedirs(features_dir, exist_ok=True)
 
+    # Count features already in terminal status at init time
+    completed_count = sum(
+        1 for f in features
+        if isinstance(f, dict) and f.get("status") in TERMINAL_STATUSES
+    )
+
     # Write pipeline.json
     pipeline_state = {
         "run_id": run_id,
@@ -241,7 +248,7 @@ def create_state_directory(state_dir, feature_list_path, features):
         "feature_list_path": abs_feature_list_path,
         "created_at": now,
         "total_features": len(features),
-        "completed_features": 0,
+        "completed_features": completed_count,
     }
     pipeline_path = os.path.join(abs_state_dir, "pipeline.json")
     with open(pipeline_path, "w", encoding="utf-8") as f:
@@ -260,9 +267,13 @@ def create_state_directory(state_dir, feature_list_path, features):
         sessions_dir = os.path.join(feature_dir, "sessions")
         os.makedirs(sessions_dir, exist_ok=True)
 
+        # Respect existing terminal status from feature-list.json
+        fl_status = feature.get("status", "pending")
+        init_status = fl_status if fl_status in TERMINAL_STATUSES else "pending"
+
         feature_status = {
             "feature_id": fid,
-            "status": "pending",
+            "status": init_status,
             "retry_count": 0,
             "max_retries": 3,
             "sessions": [],

package/bundled/dev-pipeline/scripts/update-feature-status.py

@@ -109,10 +109,15 @@ def now_iso():
     return datetime.now(timezone.utc).strftime("%Y-%m-%dT%H:%M:%SZ")
 
 
-def load_feature_status(state_dir, feature_id):
+def load_feature_status(state_dir, feature_id, feature_list_status=None):
     """Load the status.json for a feature.
 
     If the file does not exist, return a default pending status.
+
+    If feature_list_status is a terminal status (completed, failed, skipped),
+    it overrides the status field from status.json. This makes feature-list.json
+    the single source of truth for terminal statuses, while all other fields
+    (retry_count, sessions, etc.) still come from status.json.
     """
     status_path = os.path.join(
         state_dir, "features", feature_id, "status.json"
@@ -134,7 +139,7 @@ def load_feature_status(state_dir, feature_id):
     if err:
         # If we can't read it, treat as pending
         now = now_iso()
-        return {
+        data = {
             "feature_id": feature_id,
             "status": "pending",
             "retry_count": 0,
@@ -145,6 +150,9 @@ def load_feature_status(state_dir, feature_id):
             "created_at": now,
             "updated_at": now,
         }
+    # feature-list.json wins for terminal statuses
+    if feature_list_status in TERMINAL_STATUSES:
+        data["status"] = feature_list_status
     return data
 
 
@@ -303,7 +311,7 @@ def action_get_next(feature_list_data, state_dir):
         fid = feature.get("id")
         if not fid:
             continue
-        fs = load_feature_status(state_dir, fid)
+        fs = load_feature_status(state_dir, fid, feature.get("status"))
         status_map[fid] = fs.get("status", "pending")
         status_data_map[fid] = fs
 
@@ -574,7 +582,7 @@ def _format_duration(seconds):
         return "{}h{}m".format(h, m)
 
 
-def _estimate_remaining_time(features, state_dir, counts):
+def _estimate_remaining_time(features, state_dir, counts, feature_list_data=None):
     """基于已完成 Feature 的历史耗时，按 complexity 加权预估剩余时间。
 
     策略:
@@ -588,6 +596,13 @@ def _estimate_remaining_time(features, state_dir, counts):
     # complexity 权重（用于没有历史数据时的估算）
     COMPLEXITY_WEIGHT = {"low": 1.0, "medium": 2.0, "high": 4.0}
 
+    # Build feature-list status map for terminal status override
+    fl_status_map = {}
+    if feature_list_data:
+        for f in feature_list_data.get("features", []):
+            if isinstance(f, dict) and f.get("id"):
+                fl_status_map[f["id"]] = f.get("status")
+
     # 按 complexity 分组收集已完成 Feature 的耗时
     duration_by_complexity = {}  # complexity -> [duration_seconds]
     feature_complexity_map = {}  # feature_id -> complexity
@@ -608,7 +623,7 @@ def _estimate_remaining_time(features, state_dir, counts):
         fid = feature.get("id")
         if not fid:
             continue
-        fs = load_feature_status(state_dir, fid)
+        fs = load_feature_status(state_dir, fid, fl_status_map.get(fid))
         if fs.get("status") != "completed":
             continue
         duration = _calc_feature_duration(state_dir, fid)
@@ -638,7 +653,7 @@ def _estimate_remaining_time(features, state_dir, counts):
         fid = feature.get("id")
         if not fid:
             continue
-        fs = load_feature_status(state_dir, fid)
+        fs = load_feature_status(state_dir, fid, fl_status_map.get(fid))
         fstatus = fs.get("status", "pending")
         if fstatus in TERMINAL_STATUSES:
             continue
@@ -694,7 +709,7 @@ def action_status(feature_list_data, state_dir):
         fid = feature.get("id")
         if not fid:
             continue
-        fs = load_feature_status(state_dir, fid)
+        fs = load_feature_status(state_dir, fid, feature.get("status"))
         status_map[fid] = fs.get("status", "pending")
 
     for feature in features:
@@ -705,7 +720,7 @@ def action_status(feature_list_data, state_dir):
         if not fid:
             continue
 
-        fs = load_feature_status(state_dir, fid)
+        fs = load_feature_status(state_dir, fid, feature.get("status"))
         fstatus = fs.get("status", "pending")
         retry_count = fs.get("retry_count", 0)
         max_retries_val = fs.get("max_retries", 3)
@@ -797,7 +812,7 @@ def action_status(feature_list_data, state_dir):
 
     # 预估剩余时间
     est_remaining, confidence = _estimate_remaining_time(
-        features, state_dir, counts
+        features, state_dir, counts, feature_list_data
     )
 
     summary_line = "Total: {} features | Completed: {} | In Progress: {}".format(

package/bundled/skills/_metadata.json

@@ -1,5 +1,5 @@
 {
-  "version": "1.0.26",
+  "version": "1.0.28",
   "skills": {
     "prizm-kit": {
       "description": "Full-lifecycle dev toolkit. Covers spec-driven development, Prizm context docs, code quality, debugging, deployment, and knowledge management.",
@@ -197,13 +197,6 @@
       "hasAssets": false,
       "hasScripts": false
     },
-    "refactor-skill": {
-      "description": "Intelligent refactor review for existing skills with in-place vs v2 optimization and mandatory eval+graphical review.",
-      "tier": "companion",
-      "category": "Custom-skill",
-      "hasAssets": false,
-      "hasScripts": false
-    },
     "bug-planner": {
       "description": "Interactive bug planning that produces bug-fix-list.json. Supports stack traces, user reports, failed tests, log patterns, monitoring alerts.",
       "tier": "companion",
@@ -255,7 +248,6 @@
         "prizmkit-retrospective",
         "feature-workflow",
         "refactor-workflow",
-        "refactor-skill",
         "app-planner",
         "bug-planner",
         "dev-pipeline-launcher",

package/bundled/skills/dev-pipeline-launcher/SKILL.md

@@ -99,9 +99,34 @@ Detect user intent from their message, then follow the corresponding workflow:
      --action status 2>/dev/null
    ```
 
-4. **Ask user to confirm**: "Ready to launch the pipeline? It will process N features in the background."
+4. **Ask execution mode**: Present the user with a choice before launching:
+   - **(1) Background daemon (recommended)**: Pipeline runs fully detached via `launch-daemon.sh`. Survives session closure.
+   - **(2) Foreground in session**: Pipeline runs in the current session via `run.sh run`. Visible output but will stop if session times out.
+   - **(3) Manual — show commands**: Display the exact commands the user can run themselves. No execution.
 
-5. **Launch**:
+   Default to option 1 if user says "just run it" or doesn't specify.
+
+   **If option 2 (foreground)**:
+   ```bash
+   dev-pipeline/run.sh run feature-list.json
+   ```
+   Note: This will block the session. Warn user about timeout risk.
+
+   **If option 3 (manual)**: Print commands and stop. Do not execute anything.
+   ```
+   # To run in background (recommended):
+   dev-pipeline/launch-daemon.sh start feature-list.json
+
+   # To run in foreground:
+   dev-pipeline/run.sh run feature-list.json
+
+   # To check status:
+   dev-pipeline/launch-daemon.sh status
+   ```
+
+5. **Ask user to confirm**: "Ready to launch the pipeline? It will process N features in the background."
+
+6. **Launch**:
    ```bash
    dev-pipeline/launch-daemon.sh start feature-list.json
    ```
@@ -110,18 +135,18 @@ Detect user intent from their message, then follow the corresponding workflow:
    dev-pipeline/launch-daemon.sh start feature-list.json --env "SESSION_TIMEOUT=7200 MAX_RETRIES=5"
    ```
 
-6. **Verify launch**:
+7. **Verify launch**:
    ```bash
    dev-pipeline/launch-daemon.sh status
    ```
 
-7. **Start log monitoring** -- Use the Bash tool with `run_in_background: true`:
+8. **Start log monitoring** -- Use the Bash tool with `run_in_background: true`:
    ```bash
    tail -f dev-pipeline/state/pipeline-daemon.log
    ```
    This runs in background so you can continue interacting with the user.
 
-8. **Report to user**:
+9. **Report to user**:
    - Pipeline PID
    - Log file location
    - "You can ask me 'pipeline status' or 'show logs' at any time"

package/bundled/skills/feature-workflow/SKILL.md

@@ -144,12 +144,19 @@ Add new features to an existing project (incremental mode).
    Proceed? (Y/n)
    ```
 
-2. **Invoke `dev-pipeline-launcher` skill**:
+2. **Ask execution mode**: Before invoking the launcher, present the choice:
+   - **(1) Background daemon (recommended)**: Runs detached, survives session closure.
+   - **(2) Foreground in session**: Runs in current session with visible output. Stops if session times out.
+   - **(3) Manual — show commands**: Display commands only, no execution.
+
+   Pass the chosen mode to `dev-pipeline-launcher`.
+
+3. **Invoke `dev-pipeline-launcher` skill**:
    - The launcher handles all prerequisites checks
    - Starts `launch-daemon.sh` in background
    - Returns PID and log file location
 
-3. **Verify launch success**:
+4. **Verify launch success**:
    - Confirm pipeline is running
    - Record PID and log path for Phase 3
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "prizmkit",
-  "version": "1.0.26",
+  "version": "1.0.28",
   "description": "Create a new PrizmKit-powered project with clean initialization — no framework dev files, just what you need.",
   "type": "module",
   "bin": {

package/src/index.js

@@ -45,7 +45,6 @@ export async function runScaffold(directory, options) {
   const cliStatus = [
     detected.cbc ? chalk.green('cbc ✓') : chalk.gray('cbc ✗'),
     detected.claude ? chalk.green('claude ✓') : chalk.gray('claude ✗'),
-    detected.claudeInternal ? chalk.green('claude-internal ✓') : chalk.gray('claude-internal ✗'),
   ].join('  ');
   console.log(`  检测到的 CLI 工具:  ${cliStatus}`);
   console.log(`  目标目录:          ${projectRoot}`);

package/bundled/skills/refactor-skill/SKILL.md

@@ -1,371 +0,0 @@
----
-name: "refactor-skill"
-tier: companion
-description: "Intelligent refactor review for existing skills: evaluates quality, proposes in-place upgrade, and enforces eval + graphical review after changes. (project) by using the newest skill-creator standard"
----
-
-# Refactor Skill
-
-Specialized workflow for reviewing and upgrading existing skills with measurable quality gates.
-
-## When to Use
-
-Use this skill when user says:
-- "重构这个 skill", "优化技能设计", "评审并升级技能"
-- "review this skill and improve it"
-- "keep the same skill but improve quality"
-- "原地升级这个 skill" / "in-place upgrade this skill"
-
-Do NOT use when user only wants to run a pipeline immediately without changing skill design.
-
-## Core Goals
-
-1. Review current skill comprehensively and find concrete improvement points.
-2. **Default and preferred mode: in-place upgrade** of the existing skill.
-3. New-version fork (e.g., `-v2`) is **exception-only** and must be explicitly requested by user.
-4. After any modification, **must** run standardized evaluation and graphical review.
-
-## Context Readiness Gate (Mandatory)
-
-Before any refactor action, verify whether conversation context already contains:
-- target skill name/path
-- current project/workspace path
-- refactor objective and constraints (quality, speed, compatibility)
-- whether user explicitly requests a new-version fork (default is in-place upgrade)
-
-If any item is missing, do not block; gather context proactively:
-1. Read `/core/skills/_metadata.json` to locate target skill and related neighbors.
-2. Read target `SKILL.md` plus key assets/scripts under that skill.
-3. Check recent evaluation artifacts under `/.codebuddy/skill-evals/` if present.
-4. Ask only the minimum unresolved question(s).
-
-## Review Dimensions (Mandatory Rubric)
-
-Assess and score each dimension (1-5):
-
-1. **功能性 (Functionality)**
-   - Trigger clarity and routing correctness
-   - Workflow completeness and error recovery
-   - Output contract correctness (schema/format compatibility)
-
-2. **效率性 (Efficiency)**
-   - Unnecessary steps, token/time overhead
-   - Reusability of scripts/assets
-   - Fast-path design and fallback strategy
-
-3. **可维护性 (Maintainability)**
-   - Instruction structure/readability
-   - Coupling to environment and path robustness
-   - Testability and observability (artifacts, checkpoints)
-
-Output a concise review summary with:
-- strengths
-- prioritized issues (P0/P1/P2)
-- expected impact for each fix
-
-## Optimization Strategy Selection
-
-After review, apply **in-place upgrade** by default.
-
-### Default Mode — In-Place Upgrade (Required Unless Explicitly Overridden)
-Use in almost all cases:
-- skill naming and contract remain stable
-- change scope is moderate or large but compatible
-- backward compatibility is required
-
-Actions:
-1. edit existing skill files in place
-2. preserve skill name/frontmatter compatibility
-3. keep migration notes minimal
-
-### Exception Mode — New Version via `skill-creator` (Explicit User Request Only)
-Only use when user clearly asks to fork a new version (e.g., `create <skill>-v2`).
-
-Additional required checks before using exception mode:
-- user confirms they need side-by-side old/new variants
-- user accepts added maintenance cost for two versions
-
-Actions:
-1. copy current skill as baseline snapshot
-2. create `<skill-name>-v2` and apply redesign
-3. run evaluation against old version baseline
-
-## Mandatory Post-Change Validation, Review, and Optimization Loop
-
-Run this full loop after **every** refactor (default: in-place; exception: new-version fork). Do not skip.
-
-### Step 0: Freeze Refactor Scope (Input Gate)
-Capture and freeze:
-- target skill path
-- iteration id (`iteration-N`)
-- baseline type (`old-snapshot` preferred, fallback `without_skill`)
-- optimization goal for this round (quality, token, latency, or compatibility)
-
-Expected output:
-- one-line run plan: `skill + baseline + iteration + goal`
-
-### Step 1: Structural Validation (Pre-Eval)
-Validate skill structure/frontmatter and required files first.
-
-Expected output:
-- validation pass/fail result
-- blocking fix list if failed
-
-### Step 2: Execute Standardized Eval Runs (Mandatory)
-Create iteration workspace:
-- `/.codebuddy/skill-evals/<skill-name>-workspace/iteration-N/`
-
-Run both configurations for the same eval set in the same iteration:
-- `with_skill` (updated skill)
-- `baseline` (old snapshot or `without_skill`)
-
-Use multi-run strategy:
-- default: 3 runs (fast feedback)
-- release gate: 5 runs (stability check)
-
-Required artifacts per run:
-- `outputs/`
-- `timing.json`
-- `grading.json`
-- `eval_metadata.json` (per eval directory)
-
-Expected output:
-- complete run tree with paired `with_skill` vs `baseline` runs
-- no missing required artifact files
-
-### Step 3: Score, Aggregate, and Build Benchmark
-Run grading and aggregation using standardized scripts. Keep metrics comparable across iterations.
-
-Required outputs:
-- `benchmark.json`
-- `benchmark.md`
-
-Required metrics:
-- pass rate
-- duration
-- token usage
-- with_skill vs baseline delta
-- variance (`stddev`) for stability judgment
-
-Expected output:
-- benchmark summary with clear win/lose/neutral conclusion per metric
-
-### Step 4: Graphical Review (Mandatory via `generate_review`)
-Generate review UI using official `generate_review.py` (no custom viewer).
-
-Preferred modes:
-1. server mode for interactive inspection
-2. static HTML mode (`--static`) for headless fallback
-
-Expected output:
-- review entry recorded (URL or HTML path)
-- quick notes on representative good/bad runs linked to evidence
-
-### Step 5: Analyze Results and Derive Optimization Actions
-Translate benchmark + viewer evidence into prioritized actions:
-- **P0**: contract/validation breakages
-- **P1**: quality instability or high variance
-- **P2**: token/time inefficiencies
-
-For each action define:
-- root cause hypothesis
-- exact file/section to modify
-- expected metric impact
-- rollback condition
-
-Expected output:
-- actionable optimization list (not generic advice)
-
-### Step 6: Implement Targeted Improvements
-Apply only the selected actions for this iteration.
-Avoid mixing unrelated changes to keep causal attribution clear.
-
-Expected output:
-- focused diff scoped to the chosen actions
-
-### Step 7: Re-Run Evaluation and Compare Iterations
-Re-run Step 2–4 on the updated skill and compare against previous iteration.
-
-Decision rule:
-- if goals met and no regression: accept iteration
-- if partial improvement: keep gains, open next iteration with narrowed scope
-- if regression: rollback or revise hypothesis and repeat
-
-Expected output:
-- iteration verdict (`accepted` / `needs-next-iteration` / `rollback`)
-- before/after comparison table
-
-### Step 8: Close the Loop (Mandatory Delivery)
-Return:
-1. what changed
-2. measured impact (pass/time/tokens/variance deltas)
-3. viewer entry
-4. remaining risks
-5. next iteration plan (if needed)
-
-This closes the loop from **test review → evidence analysis → skill optimization → re-validation**.
-
-### Standard Command Blueprint (Project-level)
-Use the one-command review pipeline with optional grader hook:
-
-```bash
-npm run skill:review -- \
-  --workspace /abs/.codebuddy/skill-evals/<skill-name>-workspace \
-  --iteration iteration-N \
-  --skill-name <skill-name> \
-  --skill-path /abs/core/skills/<skill-name> \
-  --runs 3 \
-  --grader-cmd "python3 /abs/scripts/skill-evals/grade-eval-runs.py --workspace {workspace} --iteration {iteration} --validator /abs/core/skills/<skill-name>/scripts/validate-and-generate.py --baseline-input /abs/.codebuddy/skill-evals/<skill-name>-workspace/inputs/feature-list-existing.json"
-```
-
-Minimum expected deliverables per iteration:
-- `<workspace>/<iteration>/benchmark.json`
-- `<workspace>/<iteration>/benchmark.md`
-- `<workspace>/<iteration>/review.html`
-- optimization action list with priority and owner
-
-## Execution Notes for `skill-creator` Integration
-
-When available, follow latest `skill-creator` evaluation/viewer workflow as source of truth:
-- parallelized run spawning (with_skill + baseline)
-- assertion-based grading format compatibility
-- benchmark aggregation via official script
-- viewer generation via official script
-
-## Output Contract of This Skill
-
-After completion, return:
-1. selected mode (`in-place` by default, or `new-version` if explicitly requested) and why
-2. files changed/created
-3. review rubric scores before vs after
-4. benchmark summary (pass/time/tokens delta)
-5. graphical review entry (URL or static HTML path)
-6. remaining risks and next iteration suggestions
-
-## Error Handling
-
-- Missing target skill path: auto-discover under `/core/skills/` then confirm.
-- Missing baseline snapshot: create one before modifications.
-- Eval incomplete: mark status as blocked and list missing artifacts.
-- Viewer runtime incompatibility: switch to `--static` mode and continue.
-
-## Skill Registry Modification Guide
-
-When adding or removing skills from the framework, follow this reference checklist.
-
-### Adding a New Skill
-
-**Step 1: Create skill definition**
-```
-core/skills/<skill-name>/SKILL.md    # Required: skill definition with frontmatter
-core/skills/<skill-name>/assets/     # Optional: templates, configs, etc.
-core/skills/<skill-name>/scripts/    # Optional: executable scripts
-```
-
-**Step 2: Register in metadata**
-
-Edit `core/skills/_metadata.json`:
-```json
-{
-  "skills": {
-    "<skill-name>": {
-      "description": "Brief description of the skill",
-      "tier": "1",              // "foundation", "1", "2", or "companion"
-      "category": "core",       // "core", "quality", "devops", "debugging", "documentation", "pipeline"
-      "hasAssets": false,       // true if assets/ directory exists
-      "hasScripts": false       // true if scripts/ directory exists
-    }
-  }
-}
-```
-
-**Step 3: (Optional) Add to suite**
-
-If the skill belongs to `core` or `minimal` suite, add to `suites` section in `_metadata.json`:
-```json
-{
-  "suites": {
-    "core": {
-      "skills": ["<skill-name>", ...]
-    }
-  }
-}
-```
-
-**Step 4: Regenerate derived artifacts**
-```bash
-# Update bundled directory for npm package
-node scripts/bundle.js
-```
-
-**Step 5: Validate**
-```bash
-npm test
-# or
-node tests/validate-all.js
-```
-
-### Removing a Skill
-
-**Step 1: Delete skill directory**
-```bash
-rm -rf core/skills/<skill-name>/
-```
-
-**Step 2: Remove from metadata**
-
-Edit `core/skills/_metadata.json`:
-- Remove entry from `skills` object
-- Remove from any `suites` that reference it
-
-**Step 3: Regenerate derived artifacts**
-```bash
-node scripts/bundle.js
-```
-
-**Step 4: Validate**
-```bash
-npm test
-```
-
-### Modification Checklist Summary
-
-| File | Action | Required |
-|------|--------|----------|
-| `core/skills/<name>/SKILL.md` | Create/Delete | **Always** |
-| `core/skills/_metadata.json` → `skills` | Add/Remove entry | **Always** |
-| `core/skills/_metadata.json` → `suites` | Add/Remove from suite | If belongs to suite |
-| `core/skills/<name>/assets/` | Create/Delete | If has resources |
-| `core/skills/<name>/scripts/` | Create/Delete | If has scripts |
-| `create-prizmkit/bundled/` | Regenerate via script | Auto |
-
-### Documents That May Need Number Updates
-
-**Recommendation: Avoid hardcoding skill counts.** Use relative descriptions instead:
-- ✅ "All skills" instead of "34 skills"
-- ✅ "Core Tier 1 skills" instead of "Core Tier 1 skills (17 skills)"
-- ✅ "symlink (skills)" instead of "symlink (35 skills)"
-
-If you must include counts, maintain them in one place (`_metadata.json`) and update all references together.
-
-Files that currently contain hardcoded skill counts:
-
-| File | Current Pattern | Suggested Fix |
-|------|-----------------|---------------|
-| `README.md` | "**N Skills** covering..." | Remove number or use "Skills" |
-| `CODEBUDDY.md` | "**N Skills** — ..." | Remove number |
-| `PK-Construct-Guide.md` | "N skills — 每个 skill..." | Remove number |
-| `PK-Evolving-User-Guide.md` | "symlink (N skills)" | Use "symlink (skills)" |
-| `core/skills/prizm-kit/SKILL.md` | "## Skill Inventory (N skills)" | Use "## Skill Inventory" |
-| `core/skills/_metadata.json` | `"description": "All N skills"` | Use `"description": "All skills"` |
-
-To find hardcoded numbers:
-```bash
-grep -rn "[0-9]\+ skills\|[0-9]\+ Skills" --include="*.md" --include="*.json" .
-```
-
-## Path Rules
-
-- Prefer absolute paths in execution commands.
-- Keep path references portable in instructions when possible (e.g., `${SKILL_DIR}` for intra-skill files).
-- Never delete `.codebuddy` directory.