crewly 1.5.11 → 1.5.12

package/config/skills/agent/screenshot-compare/execute.sh

@@ -0,0 +1,182 @@
+#!/bin/bash
+# Screenshot Compare — Gemini Vision API dual-image comparison
+#
+# Sends two screenshots (e.g., iOS reference + Web implementation) to
+# Gemini Vision and returns a structured diff report.
+#
+# Usage:
+#   bash execute.sh '{"reference":"/path/to/ios-screenshot.png","target":"/path/to/web-screenshot.png"}'
+#   bash execute.sh '{"reference":"ref.png","target":"web.png","focus":"icons,layout,colors"}'
+#
+# Requires: GEMINI_API_KEY environment variable
+#
+# Output: JSON with differences categorized by type (icon, layout, color, text, image)
+
+set -euo pipefail
+
+# ── Configuration ──────────────────────────────────────────────────────────────
+MODEL="${GEMINI_MODEL:-gemini-2.5-flash-preview-05-20}"
+GEMINI_API_BASE="https://generativelanguage.googleapis.com"
+GEMINI_CONTENT_URL="${GEMINI_API_BASE}/v1beta/models/${MODEL}:generateContent"
+MAX_IMAGE_SIZE_MB=4
+
+# ── Input parsing ──────────────────────────────────────────────────────────────
+INPUT="${1:-}"
+if [ -z "$INPUT" ]; then
+  echo '{"success":false,"error":"Usage: execute.sh \u0027{\"reference\":\"/path/to/ref.png\",\"target\":\"/path/to/target.png\"}\u0027"}'
+  exit 1
+fi
+
+REFERENCE=$(echo "$INPUT" | jq -r '.reference // empty')
+TARGET=$(echo "$INPUT" | jq -r '.target // empty')
+FOCUS=$(echo "$INPUT" | jq -r '.focus // "all"')
+CONTEXT=$(echo "$INPUT" | jq -r '.context // ""')
+
+if [ -z "$REFERENCE" ] || [ -z "$TARGET" ]; then
+  echo '{"success":false,"error":"Both reference and target image paths are required"}'
+  exit 1
+fi
+
+if [ ! -f "$REFERENCE" ]; then
+  echo "{\"success\":false,\"error\":\"Reference image not found: $REFERENCE\"}"
+  exit 1
+fi
+
+if [ ! -f "$TARGET" ]; then
+  echo "{\"success\":false,\"error\":\"Target image not found: $TARGET\"}"
+  exit 1
+fi
+
+# ── API key resolution ─────────────────────────────────────────────────────────
+if [ -z "${GEMINI_API_KEY:-}" ]; then
+  # Try loading from settings via Crewly API
+  GEMINI_API_KEY=$(curl -sf "http://localhost:8787/api/settings" 2>/dev/null | jq -r '.data.apiKeys.global.gemini // empty' 2>/dev/null || true)
+fi
+
+if [ -z "${GEMINI_API_KEY:-}" ]; then
+  echo '{"success":false,"error":"GEMINI_API_KEY not set. Set via environment or Settings > API Keys."}'
+  exit 1
+fi
+
+# ── Image encoding ─────────────────────────────────────────────────────────────
+get_mime_type() {
+  local file="$1"
+  local ext="${file##*.}"
+  case "$ext" in
+    png)  echo "image/png" ;;
+    jpg|jpeg) echo "image/jpeg" ;;
+    gif)  echo "image/gif" ;;
+    webp) echo "image/webp" ;;
+    *)    echo "image/png" ;;  # default
+  esac
+}
+
+REF_MIME=$(get_mime_type "$REFERENCE")
+TGT_MIME=$(get_mime_type "$TARGET")
+REF_B64=$(base64 < "$REFERENCE" | tr -d '\n')
+TGT_B64=$(base64 < "$TARGET" | tr -d '\n')
+
+# ── Build prompt ───────────────────────────────────────────────────────────────
+FOCUS_INSTRUCTION=""
+if [ "$FOCUS" != "all" ]; then
+  FOCUS_INSTRUCTION="Focus specifically on these areas: ${FOCUS}."
+fi
+
+CONTEXT_INSTRUCTION=""
+if [ -n "$CONTEXT" ]; then
+  CONTEXT_INSTRUCTION="Additional context: ${CONTEXT}"
+fi
+
+PROMPT="You are a pixel-perfect UI comparison expert. Compare these two screenshots:
+
+IMAGE 1 (Reference — the source of truth, e.g., iOS app or design mockup)
+IMAGE 2 (Target — the implementation to verify, e.g., web app)
+
+${FOCUS_INSTRUCTION}
+${CONTEXT_INSTRUCTION}
+
+Analyze every visible difference and return a JSON array of issues found. Each issue must have:
+- \"type\": one of \"icon_missing\", \"icon_wrong\", \"layout_shift\", \"spacing\", \"color_mismatch\", \"text_mismatch\", \"font_mismatch\", \"image_missing\", \"image_wrong\", \"border_radius\", \"shadow\", \"alignment\", \"responsive\", \"other\"
+- \"severity\": \"critical\" (functionality broken), \"major\" (clearly visible), \"minor\" (subtle)
+- \"element\": which UI element is affected (e.g., \"header logo\", \"submit button\", \"nav bar\")
+- \"reference\": what it looks like in Image 1
+- \"target\": what it looks like in Image 2
+- \"suggestion\": specific CSS/code fix suggestion
+
+Return ONLY valid JSON in this format:
+{
+  \"summary\": \"Brief overall assessment\",
+  \"matchScore\": 85,
+  \"totalIssues\": 5,
+  \"issues\": [
+    {
+      \"type\": \"...\",
+      \"severity\": \"...\",
+      \"element\": \"...\",
+      \"reference\": \"...\",
+      \"target\": \"...\",
+      \"suggestion\": \"...\"
+    }
+  ]
+}"
+
+# ── API call ───────────────────────────────────────────────────────────────────
+REQUEST_BODY=$(cat <<ENDJSON
+{
+  "contents": [{
+    "parts": [
+      {"text": $(echo "$PROMPT" | jq -Rs .)},
+      {
+        "inline_data": {
+          "mime_type": "${REF_MIME}",
+          "data": "${REF_B64}"
+        }
+      },
+      {
+        "inline_data": {
+          "mime_type": "${TGT_MIME}",
+          "data": "${TGT_B64}"
+        }
+      }
+    ]
+  }],
+  "generationConfig": {
+    "temperature": 0.1,
+    "maxOutputTokens": 4096,
+    "responseMimeType": "application/json"
+  }
+}
+ENDJSON
+)
+
+TMPFILE=$(mktemp /tmp/gemini-compare-XXXXXX.json)
+echo "$REQUEST_BODY" > "$TMPFILE"
+RESPONSE=$(curl -sf -X POST \
+  "${GEMINI_CONTENT_URL}?key=${GEMINI_API_KEY}" \
+  -H "Content-Type: application/json" \
+  -d "@${TMPFILE}" 2>&1)
+rm -f "$TMPFILE"
+
+HTTP_CODE=$?
+if [ $HTTP_CODE -ne 0 ]; then
+  echo "{\"success\":false,\"error\":\"Gemini API request failed (curl exit $HTTP_CODE)\"}"
+  exit 1
+fi
+
+# ── Parse response ─────────────────────────────────────────────────────────────
+ERROR_MSG=$(echo "$RESPONSE" | jq -r '.error.message // empty' 2>/dev/null)
+if [ -n "$ERROR_MSG" ]; then
+  echo "{\"success\":false,\"error\":\"Gemini API error: $ERROR_MSG\"}"
+  exit 1
+fi
+
+# Extract the text content from Gemini response
+RESULT_TEXT=$(echo "$RESPONSE" | jq -r '.candidates[0].content.parts[0].text // empty' 2>/dev/null)
+
+if [ -z "$RESULT_TEXT" ]; then
+  echo '{"success":false,"error":"Empty response from Gemini API"}'
+  exit 1
+fi
+
+# Output: wrap the Gemini JSON result in a success envelope
+echo "{\"success\":true,\"model\":\"${MODEL}\",\"reference\":\"${REFERENCE}\",\"target\":\"${TARGET}\",\"comparison\":${RESULT_TEXT}}"

package/config/skills/agent/screenshot-compare/skill.json

@@ -0,0 +1,10 @@
+{
+  "name": "screenshot-compare",
+  "displayName": "Screenshot Compare (Gemini Vision)",
+  "description": "Compare two UI screenshots using Gemini Vision API. Returns categorized diffs with severity and fix suggestions.",
+  "category": "qa",
+  "assignableRoles": ["*"],
+  "version": "1.0.0",
+  "author": "Crewly Team",
+  "requires": ["GEMINI_API_KEY"]
+}

package/config/skills/agent/xiaoyuzhoufm-transcript/SKILL.md

@@ -0,0 +1,85 @@
+---
+name: xiaoyuzhoufm-transcript
+description: Extract and transcribe 小宇宙 (Xiaoyuzhou FM) podcast episodes. Downloads audio from episode URL and uses Gemini Audio API for transcription with speaker identification and timestamps.
+category: content
+assignableRoles:
+  - "*"
+version: "1.0.0"
+tags:
+  - podcast
+  - transcript
+  - audio
+  - gemini
+  - xiaoyuzhou
+  - chinese
+---
+
+# 小宇宙 Podcast Transcript
+
+Extract and transcribe podcast episodes from 小宇宙 (Xiaoyuzhou FM) using Gemini Audio API.
+
+## Pipeline
+
+```
+Episode URL → Fetch HTML → Extract audio URL → Download m4a → Upload to Gemini → Transcribe → Output
+```
+
+## Usage
+
+```bash
+# Transcribe from URL
+bash execute.sh '{"url":"https://www.xiaoyuzhoufm.com/episode/69bbc8ea3c625cc5ae21b461"}'
+
+# Save transcript to file
+bash execute.sh '{"url":"https://www.xiaoyuzhoufm.com/episode/...","outputFile":"./transcripts/episode.md"}'
+
+# Transcribe a local audio file (skip download)
+bash execute.sh '{"audioFile":"/path/to/podcast.m4a"}'
+
+# Specify language hint
+bash execute.sh '{"url":"...","language":"Chinese"}'
+```
+
+## Parameters
+
+| Parameter | Required | Description |
+|-----------|----------|-------------|
+| `url` | Yes* | 小宇宙 episode URL |
+| `audioFile` | Yes* | Path to local audio file (alternative to url) |
+| `outputFile` | No | Save transcript as markdown to this path |
+| `language` | No | Language hint for Gemini (default: auto-detect) |
+
+*Either `url` or `audioFile` is required.
+
+## Output Format
+
+```json
+{
+  "success": true,
+  "title": "Episode Title",
+  "url": "https://www.xiaoyuzhoufm.com/episode/...",
+  "audioUrl": "https://media.xyzcdn.net/.../audio.m4a",
+  "model": "gemini-2.5-flash-preview-05-20",
+  "transcriptLength": 12345,
+  "transcript": "**Speaker 1:** [00:00:00] Welcome to the show..."
+}
+```
+
+## Transcript Format
+
+- Speaker turns identified with `**Speaker Name:** text`
+- Timestamps at segment/topic changes: `[HH:MM:SS]`
+- Preserves original language (Chinese/English/mixed)
+
+## Prerequisites
+
+- `GEMINI_API_KEY` environment variable or configured in Settings > API Keys
+- `curl` and `jq` available in PATH
+- Internet access to xiaoyuzhoufm.com and Gemini API
+
+## Notes
+
+- Audio files are typically 30-90 minutes (20-80 MB)
+- Gemini processing takes 30-120 seconds depending on audio length
+- Maximum wait time: 5 minutes for Gemini processing
+- Temp files are auto-cleaned on exit

package/config/skills/agent/xiaoyuzhoufm-transcript/execute.sh

@@ -0,0 +1,306 @@
+#!/bin/bash
+# 小宇宙 Podcast Transcript — Extract and transcribe podcast episodes
+#
+# Pipeline: URL → HTML → audio URL → download → ffmpeg split → Gemini per-chunk → merge
+#
+# Long podcasts (>15min) are split into 10-minute chunks via ffmpeg,
+# each chunk is uploaded and transcribed separately, then merged.
+#
+# Usage:
+#   bash execute.sh '{"url":"https://www.xiaoyuzhoufm.com/episode/69bbc8ea3c625cc5ae21b461"}'
+#   bash execute.sh '{"url":"...","outputFile":"/path/to/transcript.md"}'
+#   bash execute.sh '{"audioFile":"/path/to/local.m4a"}'
+#
+# Requires: GEMINI_API_KEY, ffmpeg, curl, jq
+
+set -euo pipefail
+
+# ── Configuration ──────────────────────────────────────────────────────────────
+MODEL="${GEMINI_MODEL:-gemini-2.0-flash}"
+GEMINI_API_BASE="https://generativelanguage.googleapis.com"
+GEMINI_UPLOAD_URL="${GEMINI_API_BASE}/upload/v1beta/files"
+GEMINI_CONTENT_URL="${GEMINI_API_BASE}/v1beta/models/${MODEL}:generateContent"
+MAX_WAIT_SECONDS=300
+CHUNK_DURATION_SECONDS=600  # 10 minutes per chunk
+
+# ── Input parsing ──────────────────────────────────────────────────────────────
+INPUT="${1:-}"
+if [ -z "$INPUT" ]; then
+  echo '{"success":false,"error":"Usage: execute.sh \u0027{\"url\":\"https://www.xiaoyuzhoufm.com/episode/...\"}\u0027"}'
+  exit 1
+fi
+
+URL=$(echo "$INPUT" | jq -r '.url // empty')
+AUDIO_FILE=$(echo "$INPUT" | jq -r '.audioFile // empty')
+OUTPUT_FILE=$(echo "$INPUT" | jq -r '.outputFile // empty')
+LANGUAGE=$(echo "$INPUT" | jq -r '.language // "auto"')
+
+if [ -z "$URL" ] && [ -z "$AUDIO_FILE" ]; then
+  echo '{"success":false,"error":"Either url or audioFile is required"}'
+  exit 1
+fi
+
+# ── Dependency check ───────────────────────────────────────────────────────────
+if ! command -v ffmpeg >/dev/null 2>&1; then
+  echo '{"success":false,"error":"ffmpeg is required but not installed. Install via: brew install ffmpeg"}'
+  exit 1
+fi
+
+# ── API key resolution ─────────────────────────────────────────────────────────
+if [ -z "${GEMINI_API_KEY:-}" ]; then
+  GEMINI_API_KEY=$(curl -sf "http://localhost:8787/api/settings" 2>/dev/null \
+    | jq -r '.data.apiKeys.global.gemini // empty' 2>/dev/null || true)
+fi
+if [ -z "${GEMINI_API_KEY:-}" ]; then
+  echo '{"success":false,"error":"GEMINI_API_KEY not set. Set via environment or Settings > API Keys."}'
+  exit 1
+fi
+
+# ── Temp directory with auto-cleanup ───────────────────────────────────────────
+TMPDIR_WORK=$(mktemp -d /tmp/xiaoyuzhoufm-XXXXXX)
+cleanup() { rm -rf "$TMPDIR_WORK"; }
+trap cleanup EXIT
+
+# ── Step 1: Extract audio URL from 小宇宙 page ────────────────────────────────
+TITLE="Unknown Episode"
+AUDIO_URL=""
+
+if [ -n "$URL" ] && [ -z "$AUDIO_FILE" ]; then
+  echo '{"status":"fetching","message":"Fetching episode page..."}' >&2
+
+  HTML_FILE="${TMPDIR_WORK}/page.html"
+  curl -sL -o "$HTML_FILE" \
+    -H "User-Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36" \
+    "$URL" 2>/dev/null
+
+  if [ ! -s "$HTML_FILE" ]; then
+    echo '{"success":false,"error":"Failed to fetch episode page"}'
+    exit 1
+  fi
+
+  TITLE=$(grep -o '<title>[^<]*</title>' "$HTML_FILE" \
+    | sed 's/<title>//;s/<\/title>//' \
+    | sed 's/ - .* | 小宇宙.*//' \
+    | head -1)
+  [ -z "$TITLE" ] && TITLE="Unknown Episode"
+
+  AUDIO_URL=$(grep -oE 'https://media\.xyzcdn\.net/[^"]*\.(m4a|mp3|aac|ogg)' "$HTML_FILE" | head -1)
+  if [ -z "$AUDIO_URL" ]; then
+    AUDIO_URL=$(grep -oE 'https://[^"]*\.(m4a|mp3|aac|ogg)' "$HTML_FILE" | grep -v 'apple\|google\|spotify' | head -1)
+  fi
+  if [ -z "$AUDIO_URL" ]; then
+    echo '{"success":false,"error":"Could not find audio URL in page HTML"}'
+    exit 1
+  fi
+
+  echo "{\"status\":\"downloading\",\"message\":\"Downloading audio...\",\"title\":$(echo "$TITLE" | jq -Rs .)}" >&2
+
+  AUDIO_FILE="${TMPDIR_WORK}/audio.m4a"
+  curl -sL -o "$AUDIO_FILE" \
+    -H "User-Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36" \
+    -H "Referer: https://www.xiaoyuzhoufm.com/" \
+    "$AUDIO_URL" 2>/dev/null
+
+  if [ ! -s "$AUDIO_FILE" ]; then
+    echo '{"success":false,"error":"Failed to download audio file"}'
+    exit 1
+  fi
+  AUDIO_SIZE=$(wc -c < "$AUDIO_FILE" | tr -d ' ')
+  echo "{\"status\":\"downloaded\",\"message\":\"Audio downloaded (${AUDIO_SIZE} bytes)\"}" >&2
+fi
+
+if [ ! -f "$AUDIO_FILE" ]; then
+  echo "{\"success\":false,\"error\":\"Audio file not found: ${AUDIO_FILE}\"}"
+  exit 1
+fi
+
+# ── Step 2: Get duration and split into chunks ─────────────────────────────────
+DURATION_SECS=$(ffprobe -v quiet -show_entries format=duration -of csv=p=0 "$AUDIO_FILE" 2>/dev/null | cut -d. -f1)
+DURATION_SECS=${DURATION_SECS:-0}
+
+CHUNKS_DIR="${TMPDIR_WORK}/chunks"
+mkdir -p "$CHUNKS_DIR"
+
+if [ "$DURATION_SECS" -gt "$CHUNK_DURATION_SECONDS" ]; then
+  NUM_CHUNKS=$(( (DURATION_SECS + CHUNK_DURATION_SECONDS - 1) / CHUNK_DURATION_SECONDS ))
+  echo "{\"status\":\"splitting\",\"message\":\"Splitting ${DURATION_SECS}s audio into ${NUM_CHUNKS} chunks of ${CHUNK_DURATION_SECONDS}s...\"}" >&2
+
+  ffmpeg -i "$AUDIO_FILE" -f segment -segment_time "$CHUNK_DURATION_SECONDS" \
+    -c copy -reset_timestamps 1 \
+    "${CHUNKS_DIR}/chunk_%03d.m4a" 2>/dev/null
+
+  CHUNK_FILES=$(ls "${CHUNKS_DIR}"/chunk_*.m4a 2>/dev/null | sort)
+else
+  # Short audio — single chunk
+  cp "$AUDIO_FILE" "${CHUNKS_DIR}/chunk_000.m4a"
+  CHUNK_FILES="${CHUNKS_DIR}/chunk_000.m4a"
+  NUM_CHUNKS=1
+  echo "{\"status\":\"splitting\",\"message\":\"Short audio (${DURATION_SECS}s), no splitting needed.\"}" >&2
+fi
+
+# ── Helpers: upload + transcribe a single chunk ────────────────────────────────
+
+# Determine MIME type from file extension
+get_mime() {
+  case "${1##*.}" in
+    mp3) echo "audio/mpeg" ;; m4a) echo "audio/mp4" ;; aac) echo "audio/aac" ;;
+    ogg) echo "audio/ogg" ;; wav) echo "audio/wav" ;; *) echo "audio/mp4" ;;
+  esac
+}
+
+upload_and_wait() {
+  local filepath="$1"
+  local mime_type="$2"
+  local file_size
+  file_size=$(wc -c < "$filepath" | tr -d ' ')
+  local display_name
+  display_name=$(basename "$filepath")
+
+  # Resumable upload: init
+  local init_resp
+  init_resp=$(curl -s -D- -X POST \
+    "${GEMINI_UPLOAD_URL}?key=${GEMINI_API_KEY}" \
+    -H "X-Goog-Upload-Protocol: resumable" \
+    -H "X-Goog-Upload-Command: start" \
+    -H "X-Goog-Upload-Header-Content-Length: ${file_size}" \
+    -H "X-Goog-Upload-Header-Content-Type: ${mime_type}" \
+    -H "Content-Type: application/json" \
+    -d "{\"file\":{\"display_name\":\"${display_name}\"}}" 2>&1)
+
+  local upload_url
+  upload_url=$(echo "$init_resp" | grep -i 'x-goog-upload-url:' | sed 's/.*: //' | tr -d '\r')
+  [ -z "$upload_url" ] && { echo ""; return 1; }
+
+  # Resumable upload: send bytes
+  local upload_resp
+  upload_resp=$(curl -s -X POST "$upload_url" \
+    -H "X-Goog-Upload-Offset: 0" \
+    -H "X-Goog-Upload-Command: upload, finalize" \
+    -H "Content-Type: ${mime_type}" \
+    --data-binary "@${filepath}" 2>&1)
+
+  local file_uri file_name
+  file_uri=$(echo "$upload_resp" | jq -r '.file.uri // empty' 2>/dev/null)
+  file_name=$(echo "$upload_resp" | jq -r '.file.name // empty' 2>/dev/null)
+  [ -z "$file_uri" ] && { echo ""; return 1; }
+
+  # Poll until ACTIVE
+  local state="PROCESSING" waited=0
+  while [ "$state" = "PROCESSING" ] && [ "$waited" -lt "$MAX_WAIT_SECONDS" ]; do
+    sleep 3
+    waited=$((waited + 3))
+    state=$(curl -s "${GEMINI_API_BASE}/v1beta/${file_name}?key=${GEMINI_API_KEY}" \
+      | jq -r '.state // "PROCESSING"' 2>/dev/null)
+  done
+
+  [ "$state" != "ACTIVE" ] && { echo ""; return 1; }
+  echo "$file_uri"
+}
+
+transcribe_chunk() {
+  local file_uri="$1"
+  local mime_type="$2"
+  local chunk_index="$3"
+  local total_chunks="$4"
+
+  local lang_hint=""
+  [ "$LANGUAGE" != "auto" ] && lang_hint="The audio is in ${LANGUAGE}. "
+
+  local prompt="${lang_hint}Transcribe this audio segment (chunk $((chunk_index+1)) of ${total_chunks}). Output the full transcript preserving speaker turns. Format:
+
+**Speaker Name/Label:** text spoken...
+
+Use Speaker 1, Speaker 2 etc if names unknown. Maintain original language. Add [MM:SS] timestamps at topic changes (relative to this chunk start)."
+
+  local req_file="${TMPDIR_WORK}/req_${chunk_index}.json"
+  cat > "$req_file" <<ENDJSON
+{
+  "contents": [{"parts": [
+    {"text": $(echo "$prompt" | jq -Rs .)},
+    {"file_data": {"mime_type": "${mime_type}", "file_uri": "${file_uri}"}}
+  ]}],
+  "generationConfig": {"temperature": 0.1, "maxOutputTokens": 32768}
+}
+ENDJSON
+
+  local resp
+  resp=$(curl -s --max-time 300 -X POST \
+    "${GEMINI_CONTENT_URL}?key=${GEMINI_API_KEY}" \
+    -H "Content-Type: application/json" \
+    -d "@${req_file}" 2>&1)
+
+  local err_msg
+  err_msg=$(echo "$resp" | jq -r '.error.message // empty' 2>/dev/null)
+  [ -n "$err_msg" ] && { echo "[Transcription error: ${err_msg}]"; return 0; }
+
+  echo "$resp" | jq -r '.candidates[0].content.parts[0].text // "[No transcript for this chunk]"' 2>/dev/null
+}
+
+# ── Step 3: Upload and transcribe each chunk ───────────────────────────────────
+FULL_TRANSCRIPT=""
+CHUNK_INDEX=0
+TOTAL_CHUNKS=$(echo "$CHUNK_FILES" | wc -w | tr -d ' ')
+
+for chunk_file in $CHUNK_FILES; do
+  CHUNK_INDEX_DISPLAY=$((CHUNK_INDEX + 1))
+  echo "{\"status\":\"transcribing\",\"message\":\"Chunk ${CHUNK_INDEX_DISPLAY}/${TOTAL_CHUNKS}: uploading...\"}" >&2
+
+  MIME=$(get_mime "$chunk_file")
+  FILE_URI=$(upload_and_wait "$chunk_file" "$MIME")
+
+  if [ -z "$FILE_URI" ]; then
+    echo "{\"status\":\"error\",\"message\":\"Chunk ${CHUNK_INDEX_DISPLAY} upload/processing failed, skipping\"}" >&2
+    FULL_TRANSCRIPT="${FULL_TRANSCRIPT}
+
+---
+[Chunk ${CHUNK_INDEX_DISPLAY} failed to process]
+---
+"
+    CHUNK_INDEX=$((CHUNK_INDEX + 1))
+    continue
+  fi
+
+  echo "{\"status\":\"transcribing\",\"message\":\"Chunk ${CHUNK_INDEX_DISPLAY}/${TOTAL_CHUNKS}: transcribing...\"}" >&2
+  CHUNK_TEXT=$(transcribe_chunk "$FILE_URI" "$MIME" "$CHUNK_INDEX" "$TOTAL_CHUNKS")
+
+  # Add time offset header for multi-chunk
+  if [ "$TOTAL_CHUNKS" -gt 1 ]; then
+    OFFSET_MINS=$(( CHUNK_INDEX * CHUNK_DURATION_SECONDS / 60 ))
+    FULL_TRANSCRIPT="${FULL_TRANSCRIPT}
+
+--- Segment ${CHUNK_INDEX_DISPLAY} (from ${OFFSET_MINS}:00) ---
+
+${CHUNK_TEXT}"
+  else
+    FULL_TRANSCRIPT="$CHUNK_TEXT"
+  fi
+
+  CHUNK_INDEX=$((CHUNK_INDEX + 1))
+  echo "{\"status\":\"transcribing\",\"message\":\"Chunk ${CHUNK_INDEX_DISPLAY}/${TOTAL_CHUNKS}: done.\"}" >&2
+done
+
+if [ -z "$FULL_TRANSCRIPT" ]; then
+  echo '{"success":false,"error":"All chunks failed to transcribe"}'
+  exit 1
+fi
+
+# ── Step 4: Output ─────────────────────────────────────────────────────────────
+if [ -n "$OUTPUT_FILE" ]; then
+  mkdir -p "$(dirname "$OUTPUT_FILE")"
+  cat > "$OUTPUT_FILE" <<ENDMD
+# ${TITLE}
+
+> Source: ${URL:-local file}
+> Transcribed: $(date -u +"%Y-%m-%dT%H:%M:%SZ")
+> Model: ${MODEL}
+> Duration: ${DURATION_SECS}s (${TOTAL_CHUNKS} chunks)
+
+---
+
+${FULL_TRANSCRIPT}
+ENDMD
+  echo "{\"status\":\"saved\",\"message\":\"Transcript saved to ${OUTPUT_FILE}\"}" >&2
+fi
+
+TRANSCRIPT_LENGTH=${#FULL_TRANSCRIPT}
+echo "{\"success\":true,\"title\":$(echo "$TITLE" | jq -Rs .),\"url\":$(echo "${URL:-}" | jq -Rs .),\"audioUrl\":$(echo "${AUDIO_URL:-}" | jq -Rs .),\"model\":\"${MODEL}\",\"duration\":${DURATION_SECS},\"chunks\":${TOTAL_CHUNKS},\"transcriptLength\":${TRANSCRIPT_LENGTH},\"transcript\":$(echo "$FULL_TRANSCRIPT" | jq -Rs .)}"

package/config/skills/agent/xiaoyuzhoufm-transcript/skill.json

@@ -0,0 +1,10 @@
+{
+  "name": "xiaoyuzhoufm-transcript",
+  "displayName": "小宇宙 Podcast Transcript",
+  "description": "Extract and transcribe 小宇宙 podcast episodes using Gemini Audio API. Supports speaker identification and timestamps.",
+  "category": "content",
+  "assignableRoles": ["*"],
+  "version": "1.0.0",
+  "author": "Crewly Team",
+  "requires": ["GEMINI_API_KEY"]
+}

package/config/skills/orchestrator/cancel-cron/SKILL.md

@@ -0,0 +1,44 @@
+---
+name: Cancel Cron Task
+description: Permanently delete a cron task.
+version: 1.0.0
+category: scheduling
+skillType: claude-skill
+assignableRoles:
+  - orchestrator
+triggers:
+  - cancel cron
+  - delete cron
+  - remove cron task
+tags:
+  - cron
+  - scheduling
+  - cleanup
+execution:
+  type: script
+  script:
+    file: execute.sh
+    interpreter: bash
+    timeoutMs: 15000
+---
+
+# Cancel Cron Task
+
+Permanently delete a cron task from the Crewly backend.
+
+## Usage
+
+```bash
+bash {{ORCHESTRATOR_SKILLS_PATH}}/cancel-cron/execute.sh '{"id":"cron-abc123"}'
+```
+
+## Parameters
+
+| Parameter | Required | Description |
+|-----------|----------|-------------|
+| `id` | Yes | The cron task ID to delete |
+
+## Notes
+
+- This permanently removes the cron task. To temporarily stop it, use `update-cron` with `{"enabled": false}` instead.
+- Only the orchestrator or the user can cancel cron tasks.