crewly 1.1.2 → 1.2.0

package/config/skills/agent/content-writer/instructions.md

@@ -0,0 +1,124 @@
+# Content Writer
+
+Generate platform-specific content drafts with brand voice, tone, and format guidelines. Manages the full lifecycle from writing brief to saved draft.
+
+## Workflow
+
+```
+1. Generate writing brief     → execute.sh '{"action":"draft",...}'
+2. Write content using brief  → Agent uses LLM to write
+3. Save completed draft       → execute.sh '{"action":"save",...}'
+4. Add to content calendar    → content-calendar skill
+5. Steve reviews + publishes  → Manual
+```
+
+## Actions
+
+### `draft` — Generate a writing brief
+
+Returns structured guidelines for writing a specific piece of content. The agent then uses these to generate the actual text.
+
+| Parameter | Required | Description |
+|-----------|----------|-------------|
+| `action` | Yes | `"draft"` |
+| `topic` | Yes | What to write about |
+| `platform` | Yes | `x-thread`, `x-single`, `linkedin`, `xiaohongshu`, `substack`, `youtube-desc`, `blog` |
+| `line` | No | `crewly` (brand) or `personal` (Steve). Default: `crewly` |
+| `tone` | No | `professional`, `casual`, `technical`, `inspiring`, `provocative`, `educational`. Default: `professional` |
+| `length` | No | `short`, `medium`, `long`. Default: `medium` |
+| `audience` | No | Custom audience description (overrides default) |
+| `context` | No | Additional context: reference data, trend info, competitor analysis |
+| `references` | No | URLs or docs to reference |
+| `cta` | No | Specific call-to-action to include |
+
+### `save` — Save a completed draft
+
+| Parameter | Required | Description |
+|-----------|----------|-------------|
+| `action` | Yes | `"save"` |
+| `title` | Yes | Content title |
+| `platform` | Yes | Platform name |
+| `content` | Yes | The full content text (markdown) |
+| `line` | No | Content line (default: crewly) |
+| `draftId` | No | Draft ID from the brief (for traceability) |
+| `calendarId` | No | Content calendar entry ID to link |
+
+### `get` — Read a saved draft
+
+| Parameter | Required | Description |
+|-----------|----------|-------------|
+| `action` | Yes | `"get"` |
+| `filePath` | Yes* | Path to the draft file |
+| `draftId` | Yes* | Or search by draft ID |
+
+*One of filePath or draftId required.
+
+### `list` — List saved drafts
+
+| Parameter | Required | Description |
+|-----------|----------|-------------|
+| `action` | Yes | `"list"` |
+| `platform` | No | Filter by platform |
+| `limit` | No | Max results (default: 20) |
+
+## Examples
+
+### Generate a writing brief for an X thread
+```bash
+bash execute.sh '{"action":"draft","topic":"AI agent security - what we learned from n8n 8 CVEs","platform":"x-thread","line":"crewly","tone":"provocative","context":"n8n disclosed 8 critical CVEs on 2/25/2026 including RCE via sandbox escape. Crewly uses PTY isolation."}'
+```
+
+### Generate a brief for a Xiaohongshu post
+```bash
+bash execute.sh '{"action":"draft","topic":"My AI team shipped while I slept - week 1 report","platform":"xiaohongshu","line":"personal","tone":"casual"}'
+```
+
+### Save a completed draft
+```bash
+bash execute.sh '{"action":"save","title":"AI agent security thread","platform":"x-thread","line":"crewly","content":"[1/5] n8n just disclosed 8 critical CVEs...\n\n[2/5] The scariest one?...","projectPath":"/path/to/project"}'
+```
+
+### List all LinkedIn drafts
+```bash
+bash execute.sh '{"action":"list","platform":"linkedin","projectPath":"/path/to/project"}'
+```
+
+## Platform Writing Tips
+
+### X Thread Best Practices
+- Hook tweet is everything — if they don't stop scrolling, nothing else matters
+- Each tweet = one idea, one screenshot, or one data point
+- Use numbers: "3 things I learned" > "Things I learned"
+- End with a thread-pull: "If this was useful, follow me for more"
+- Steve's top performers on Xiaohongshu all had strong visual hooks — apply same principle
+
+### LinkedIn Best Practices
+- First line shows in feed preview — make it impossible to not click "see more"
+- "I" stories outperform "You should" advice
+- Specific > general: "We saved 47 hours/week" > "We saved a lot of time"
+- Post between 8-10am EST Tuesday-Thursday for max reach
+
+### Xiaohongshu Best Practices (from Steve's data)
+- Video outperforms image-text (20 videos vs 12 image-text in Steve's archive)
+- Top tags to use: 创业MVP, 职场smalltalk, 用好ai拿捏职场, vibecoding
+- Steve's top post (201 likes): "Claude Code牛马工厂" — visual + relatable + specific tool
+- Social/community posts (Google coffee chat: 164 likes) also perform well
+- Include 8+ hashtags for discovery
+
+### Substack Best Practices
+- Subject line = open rate. Test: numbers, questions, or contrarian takes
+- Personal stories in the opening paragraph build connection
+- Include 1 actionable takeaway readers can use this week
+- Cross-promote: mention X thread, YouTube video in each issue
+
+## Brand Voice Quick Reference
+
+### Crewly Brand
+- **Do:** Use specific metrics, show product screenshots, reference real use cases
+- **Don't:** Claim "revolutionary" or "game-changing", oversell features not yet built
+- **Signature phrases:** "AI Team", "Ready in Days", "Quality Gates", "Live Terminal"
+
+### Steve Personal
+- **Do:** Share real numbers, mention Google/side projects, be vulnerable about failures
+- **Don't:** Lecture, be preachy, use generic "hustle" motivation
+- **Signature phrases:** "一人公司", "下班2小时", "留在牌桌上", "Build in Public"

package/config/skills/agent/content-writer/skill.json

@@ -0,0 +1,22 @@
+{
+  "id": "content-writer",
+  "name": "Content Writer",
+  "description": "Generate platform-specific writing briefs and manage content drafts. Supports X threads, LinkedIn posts, Xiaohongshu notes, Substack newsletters, YouTube descriptions, and blog posts. Includes brand voice presets for Crewly and Steve's personal content.",
+  "category": "content",
+  "skillType": "claude-skill",
+  "promptFile": "instructions.md",
+  "execution": {
+    "type": "script",
+    "script": {
+      "file": "execute.sh",
+      "interpreter": "bash",
+      "timeoutMs": 15000
+    }
+  },
+  "assignableRoles": ["content-strategist", "product-manager", "generalist", "designer"],
+  "triggers": ["write content", "create post", "draft content", "write thread", "write article", "content draft"],
+  "tags": ["content", "writing", "draft", "social-media", "marketing", "x-thread", "linkedin", "xiaohongshu", "substack", "youtube", "blog"],
+  "version": "1.0.0",
+  "author": "Luna (Content Strategist)",
+  "license": "MIT"
+}

package/config/skills/agent/core/generate-pdf/execute.sh

@@ -0,0 +1,88 @@
+#!/bin/bash
+# Generate PDF from Markdown with CJK font support
+# Uses pandoc (md→html) + weasyprint (html→pdf)
+set -euo pipefail
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+source "${SCRIPT_DIR}/../../_common/lib.sh"
+
+INPUT="${1:-}"
+[ -z "$INPUT" ] && error_exit "Usage: execute.sh '{\"input\":\"/tmp/report.md\",\"output\":\"/tmp/report.pdf\",\"title\":\"My Report\"}'"
+
+MD_PATH=$(echo "$INPUT" | jq -r '.input // empty')
+PDF_PATH=$(echo "$INPUT" | jq -r '.output // empty')
+TITLE=$(echo "$INPUT" | jq -r '.title // empty')
+
+require_param "input" "$MD_PATH"
+
+# Default output: same path with .pdf extension
+if [ -z "$PDF_PATH" ]; then
+  PDF_PATH="${MD_PATH%.md}.pdf"
+fi
+
+# Intermediate HTML path
+HTML_PATH="${PDF_PATH%.pdf}.html"
+
+# Default title from filename
+if [ -z "$TITLE" ]; then
+  TITLE=$(basename "$MD_PATH" .md | tr '-' ' ' | tr '_' ' ')
+fi
+
+# Ensure the CJK CSS header exists
+CJK_HEADER="/tmp/cjk-header.html"
+if [ ! -f "$CJK_HEADER" ]; then
+  cat > "$CJK_HEADER" << 'CSSEOF'
+<style>
+body {
+  font-family: -apple-system, BlinkMacSystemFont, "PingFang SC", "Hiragino Sans GB", "Noto Sans CJK SC", "Microsoft YaHei", "WenQuanYi Micro Hei", sans-serif;
+  max-width: 900px;
+  margin: 0 auto;
+  padding: 2rem;
+  line-height: 1.8;
+  color: #1a1a1a;
+  font-size: 14px;
+}
+h1, h2, h3, h4 {
+  font-family: -apple-system, "PingFang SC", "Hiragino Sans GB", sans-serif;
+  margin-top: 1.5em;
+  color: #111;
+}
+h1 { font-size: 1.8em; border-bottom: 2px solid #333; padding-bottom: 0.3em; }
+h2 { font-size: 1.4em; border-bottom: 1px solid #ccc; padding-bottom: 0.2em; }
+h3 { font-size: 1.15em; }
+table { border-collapse: collapse; width: 100%; margin: 1em 0; font-size: 13px; }
+th, td { border: 1px solid #ddd; padding: 8px 12px; text-align: left; }
+th { background: #f5f5f5; font-weight: 600; }
+tr:nth-child(even) { background: #fafafa; }
+blockquote { border-left: 4px solid #4a90d9; margin: 1em 0; padding: 0.5em 1em; background: #f0f6ff; color: #333; }
+code { background: #f4f4f4; padding: 2px 5px; border-radius: 3px; font-size: 0.9em; }
+hr { border: none; border-top: 1px solid #ddd; margin: 2em 0; }
+strong { color: #111; }
+</style>
+CSSEOF
+fi
+
+# Check dependencies
+if ! command -v pandoc &>/dev/null; then
+  error_exit "pandoc is not installed. Run: brew install pandoc"
+fi
+if ! command -v weasyprint &>/dev/null; then
+  error_exit "weasyprint is not installed. Run: brew install weasyprint"
+fi
+
+# Step 1: Markdown → HTML (with CJK CSS header)
+pandoc "$MD_PATH" -o "$HTML_PATH" \
+  --standalone \
+  --metadata title="$TITLE" \
+  --include-in-header="$CJK_HEADER" 2>/dev/null
+
+# Step 2: HTML → PDF (weasyprint handles CJK fonts correctly)
+weasyprint "$HTML_PATH" "$PDF_PATH" 2>/dev/null
+
+# Get file size
+PDF_SIZE=$(wc -c < "$PDF_PATH" | tr -d ' ')
+
+# Clean up intermediate HTML
+rm -f "$HTML_PATH"
+
+# Output result
+echo "{\"success\":true,\"pdf\":\"${PDF_PATH}\",\"size\":${PDF_SIZE}}"

package/config/skills/agent/core/generate-pdf/instructions.md

@@ -0,0 +1,46 @@
+# Generate PDF
+
+Convert a Markdown file to a professionally styled PDF with full CJK (Chinese/Japanese/Korean) font support.
+
+## Pipeline
+
+1. **Markdown → HTML**: Uses `pandoc` with `--standalone` and an inline CJK-compatible CSS header
+2. **HTML → PDF**: Uses `weasyprint` (NOT Chrome headless, which cannot render CJK characters)
+
+## Parameters
+
+| Parameter | Required | Description |
+|-----------|----------|-------------|
+| `input` | Yes | Path to the input Markdown file |
+| `output` | No | Path for the output PDF file (defaults to input path with `.pdf` extension) |
+| `title` | No | Document title for the HTML metadata |
+
+## Example
+
+```bash
+bash config/skills/agent/core/generate-pdf/execute.sh '{"input":"/tmp/report.md","output":"/tmp/report.pdf","title":"My Report"}'
+```
+
+### Minimal usage (auto-generates output path)
+
+```bash
+bash config/skills/agent/core/generate-pdf/execute.sh '{"input":"/tmp/report.md"}'
+```
+
+This produces `/tmp/report.pdf`.
+
+## Important Notes
+
+- **Always use this skill** instead of manually running pandoc + Chrome headless
+- **Chrome headless cannot render Chinese/CJK characters** — they appear blank in the PDF
+- **weasyprint** properly embeds CJK fonts (PingFang SC, Hiragino Sans GB, STHeiti) on macOS
+- The skill auto-generates a CSS header with CJK font declarations if one doesn't exist
+- Output is a clean, styled PDF suitable for sharing via Slack or email
+
+## Output
+
+JSON with the path to the generated PDF file:
+
+```json
+{"success": true, "pdf": "/tmp/report.pdf", "size": 364024}
+```

package/config/skills/agent/core/generate-pdf/skill.json

@@ -0,0 +1,20 @@
+{
+  "id": "agent-generate-pdf",
+  "name": "Generate PDF",
+  "description": "Convert a Markdown file to a styled PDF with full CJK (Chinese/Japanese/Korean) font support. Uses pandoc for Markdown→HTML and weasyprint for HTML→PDF rendering.",
+  "category": "document",
+  "skillType": "claude-skill",
+  "promptFile": "instructions.md",
+  "execution": {
+    "type": "script",
+    "script": {
+      "file": "execute.sh",
+      "interpreter": "bash",
+      "timeoutMs": 60000
+    }
+  },
+  "assignableRoles": ["developer", "qa", "tpm", "designer", "frontend-developer", "backend-developer", "fullstack-dev", "qa-engineer", "product-manager", "architect", "generalist", "sales", "support", "content-strategist"],
+  "triggers": ["generate pdf", "convert to pdf", "create pdf", "markdown to pdf", "make pdf"],
+  "tags": ["pdf", "document", "markdown", "pandoc", "weasyprint", "cjk"],
+  "version": "1.0.0"
+}

package/config/skills/agent/core/report-status/execute.sh

@@ -33,3 +33,9 @@ if [ "$STATUS" = "done" ] && [ -n "$TASK_PATH" ]; then
     '{taskPath: $taskPath, sessionName: $sessionName}')
   api_call POST "/task-management/complete" "$COMPLETE_BODY" || true
 fi
+
+# Auto-complete tracked tasks when status is done
+if [ "$STATUS" = "done" ]; then
+  SESSION_BODY=$(jq -n --arg sessionName "$SESSION_NAME" '{sessionName: $sessionName}')
+  api_call POST "/task-management/complete-by-session" "$SESSION_BODY" || true
+fi

package/config/skills/agent/trend-monitor/execute.sh

@@ -0,0 +1,211 @@
+#!/bin/bash
+# Trend Monitor — save, read, and manage trend data from browser scans
+set -euo pipefail
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+source "${SCRIPT_DIR}/../_common/lib.sh"
+
+INPUT="${1:-}"
+[ -z "$INPUT" ] && error_exit "Usage: execute.sh '{\"action\":\"save|list|latest|suggest\",\"projectPath\":\"/path/to/project\",...}'"
+
+ACTION=$(echo "$INPUT" | jq -r '.action // empty')
+PROJECT_PATH=$(echo "$INPUT" | jq -r '.projectPath // empty')
+
+require_param "action" "$ACTION"
+
+# Resolve trends directory
+if [ -n "$PROJECT_PATH" ]; then
+  TRENDS_DIR="${PROJECT_PATH}/.crewly/content/trends"
+else
+  TRENDS_DIR="${HOME}/.crewly/content/trends"
+fi
+mkdir -p "$TRENDS_DIR"
+
+case "$ACTION" in
+
+  # ─────────────────────────────────────────────
+  # SAVE: Store a batch of trends from a browser scan
+  # ─────────────────────────────────────────────
+  save)
+    SOURCE=$(echo "$INPUT" | jq -r '.source // empty')
+    TRENDS=$(echo "$INPUT" | jq -r '.trends // empty')
+
+    require_param "source" "$SOURCE"
+    require_param "trends" "$TRENDS"
+
+    # Validate source
+    case "$SOURCE" in
+      x-trending|google-trends|hackernews|producthunt|reddit|github-trending|custom) ;;
+      *) error_exit "Invalid source: $SOURCE. Valid: x-trending, google-trends, hackernews, producthunt, reddit, github-trending, custom" ;;
+    esac
+
+    # Validate trends is a JSON array
+    if ! echo "$TRENDS" | jq 'type == "array"' 2>/dev/null | grep -q true; then
+      error_exit "trends must be a JSON array of objects"
+    fi
+
+    NOW=$(date -u +%Y-%m-%dT%H:%M:%SZ)
+    TODAY=$(date -u +%Y-%m-%d)
+    SCAN_ID="scan-$(date +%s)-$((RANDOM % 1000))"
+
+    # Add metadata to each trend
+    ENRICHED=$(echo "$TRENDS" | jq --arg source "$SOURCE" --arg scanId "$SCAN_ID" --arg ts "$NOW" \
+      '[.[] | . + {"source": $source, "scanId": $scanId, "scannedAt": $ts}]')
+
+    # Count items
+    COUNT=$(echo "$ENRICHED" | jq 'length')
+
+    # Save to date-based file
+    SCAN_FILE="${TRENDS_DIR}/${TODAY}-${SOURCE}.json"
+
+    if [ -f "$SCAN_FILE" ]; then
+      # Append to existing file
+      EXISTING=$(cat "$SCAN_FILE")
+      MERGED=$(jq -n --argjson existing "$EXISTING" --argjson new "$ENRICHED" \
+        '{"scans": ($existing.scans + [{"scanId": $new[0].scanId, "scannedAt": $new[0].scannedAt, "count": ($new | length), "items": $new}])}')
+      echo "$MERGED" > "$SCAN_FILE"
+    else
+      # Create new file
+      jq -n --argjson items "$ENRICHED" --arg source "$SOURCE" --arg date "$TODAY" \
+        '{"source": $source, "date": $date, "scans": [{"scanId": $items[0].scanId, "scannedAt": $items[0].scannedAt, "count": ($items | length), "items": $items}]}' > "$SCAN_FILE"
+    fi
+
+    jq -n \
+      --arg scanId "$SCAN_ID" \
+      --arg source "$SOURCE" \
+      --argjson count "$COUNT" \
+      --arg file "$SCAN_FILE" \
+      --arg scannedAt "$NOW" \
+      '{"success":true,"action":"save","scanId":$scanId,"source":$source,"count":$count,"file":$file,"scannedAt":$scannedAt}'
+    ;;
+
+  # ─────────────────────────────────────────────
+  # LIST: List available trend scans
+  # ─────────────────────────────────────────────
+  list)
+    FILTER_SOURCE=$(echo "$INPUT" | jq -r '.source // empty')
+    FILTER_DATE=$(echo "$INPUT" | jq -r '.date // empty')
+    LIMIT=$(echo "$INPUT" | jq -r '.limit // "10"')
+
+    FILES="[]"
+    for f in "$TRENDS_DIR"/*.json; do
+      [ -f "$f" ] || continue
+      BASENAME=$(basename "$f" .json)
+      FILE_DATE=$(echo "$BASENAME" | cut -d'-' -f1-3)
+      FILE_SOURCE=$(echo "$BASENAME" | cut -d'-' -f4-)
+
+      # Apply filters
+      if [ -n "$FILTER_SOURCE" ] && [ "$FILE_SOURCE" != "$FILTER_SOURCE" ]; then
+        continue
+      fi
+      if [ -n "$FILTER_DATE" ] && [ "$FILE_DATE" != "$FILTER_DATE" ]; then
+        continue
+      fi
+
+      SCAN_COUNT=$(jq '.scans | length' "$f" 2>/dev/null || echo "0")
+      TOTAL_ITEMS=$(jq '[.scans[].count] | add // 0' "$f" 2>/dev/null || echo "0")
+      LAST_SCAN=$(jq -r '.scans[-1].scannedAt // "unknown"' "$f" 2>/dev/null || echo "unknown")
+
+      FILES=$(echo "$FILES" | jq \
+        --arg file "$f" \
+        --arg date "$FILE_DATE" \
+        --arg source "$FILE_SOURCE" \
+        --argjson scanCount "$SCAN_COUNT" \
+        --argjson totalItems "$TOTAL_ITEMS" \
+        --arg lastScan "$LAST_SCAN" \
+        '. + [{"file":$file,"date":$date,"source":$source,"scanCount":$scanCount,"totalItems":$totalItems,"lastScan":$lastScan}]')
+    done
+
+    # Sort by date descending and limit
+    FILES=$(echo "$FILES" | jq --argjson limit "$LIMIT" 'sort_by(.date) | reverse | .[:$limit]')
+    COUNT=$(echo "$FILES" | jq 'length')
+
+    jq -n --argjson files "$FILES" --argjson count "$COUNT" \
+      '{"success":true,"action":"list","count":$count,"files":$files}'
+    ;;
+
+  # ─────────────────────────────────────────────
+  # LATEST: Get the latest trends from a specific source
+  # ─────────────────────────────────────────────
+  latest)
+    FILTER_SOURCE=$(echo "$INPUT" | jq -r '.source // empty')
+    LIMIT=$(echo "$INPUT" | jq -r '.limit // "20"')
+
+    # Find most recent file(s)
+    ALL_ITEMS="[]"
+
+    for f in $(ls -t "$TRENDS_DIR"/*.json 2>/dev/null | head -5); do
+      [ -f "$f" ] || continue
+      FILE_SOURCE=$(basename "$f" .json | cut -d'-' -f4-)
+
+      if [ -n "$FILTER_SOURCE" ] && [ "$FILE_SOURCE" != "$FILTER_SOURCE" ]; then
+        continue
+      fi
+
+      # Get items from the latest scan in each file
+      ITEMS=$(jq '.scans[-1].items // []' "$f" 2>/dev/null || echo "[]")
+      ALL_ITEMS=$(jq -n --argjson a "$ALL_ITEMS" --argjson b "$ITEMS" '$a + $b')
+    done
+
+    # Sort by relevance score if available, limit
+    ALL_ITEMS=$(echo "$ALL_ITEMS" | jq --argjson limit "$LIMIT" \
+      'sort_by(.relevanceScore // 0) | reverse | .[:$limit]')
+    COUNT=$(echo "$ALL_ITEMS" | jq 'length')
+
+    jq -n --argjson items "$ALL_ITEMS" --argjson count "$COUNT" \
+      '{"success":true,"action":"latest","count":$count,"items":$items}'
+    ;;
+
+  # ─────────────────────────────────────────────
+  # SUGGEST: Generate topic suggestions from recent trends
+  # ─────────────────────────────────────────────
+  suggest)
+    CONTENT_LINE=$(echo "$INPUT" | jq -r '.line // "crewly"')
+    LIMIT=$(echo "$INPUT" | jq -r '.limit // "5"')
+
+    # Gather all recent trend items (last 3 days)
+    ALL_ITEMS="[]"
+    for f in $(ls -t "$TRENDS_DIR"/*.json 2>/dev/null | head -10); do
+      [ -f "$f" ] || continue
+      ITEMS=$(jq '.scans[-1].items // []' "$f" 2>/dev/null || echo "[]")
+      ALL_ITEMS=$(jq -n --argjson a "$ALL_ITEMS" --argjson b "$ITEMS" '$a + $b')
+    done
+
+    COUNT=$(echo "$ALL_ITEMS" | jq 'length')
+
+    if [ "$COUNT" -eq 0 ]; then
+      jq -n '{"success":true,"action":"suggest","suggestions":[],"message":"No trend data available. Run a browser scan first using the instructions in instructions.md."}'
+      exit 0
+    fi
+
+    # Filter for AI-related trends
+    AI_ITEMS=$(echo "$ALL_ITEMS" | jq '[.[] | select(
+      (.title // "" | test("(?i)ai|agent|llm|gpt|claude|gemini|automation|saas|startup")) or
+      (.relevanceScore // 0) >= 7
+    )]')
+
+    AI_COUNT=$(echo "$AI_ITEMS" | jq 'length')
+
+    # Build suggestion context
+    TOPICS=$(echo "$AI_ITEMS" | jq --argjson limit "$LIMIT" \
+      '[.[:$limit] | .[] | {title: .title, source: .source, url: (.url // ""), relevanceScore: (.relevanceScore // 0)}]')
+
+    jq -n \
+      --argjson allTrends "$COUNT" \
+      --argjson aiRelevant "$AI_COUNT" \
+      --argjson topics "$TOPICS" \
+      --arg line "$CONTENT_LINE" \
+      '{
+        "success": true,
+        "action": "suggest",
+        "totalTrendsScanned": $allTrends,
+        "aiRelevantTrends": $aiRelevant,
+        "contentLine": $line,
+        "suggestedTopics": $topics,
+        "instruction": "Use these trending topics as input for content-writer skill. Match with your content calendar and brand voice."
+      }'
+    ;;
+
+  *)
+    error_exit "Unknown action: $ACTION. Valid: save, list, latest, suggest"
+    ;;
+esac