review

name	review
description	Morning batch review of discovered content ideas. Presents top ideas from data/content.db for KEEP/SKIP/STAR decisions via CLI. Designed for Robin's morning session: one focused pass through the idea backlog to select what gets turned into content today. Trigger phrases: /review, "morning review", "show me ideas", or "what should I post about".
allowed-tools	["Read","Write","Bash","Glob","Grep","AskUserQuestion"]

Review -- Morning Batch Content Review

Interactive morning review workflow for KEEP/SKIP/STAR decisions on discovered content ideas. One focused pass through the backlog, ~5 minutes, then hand off to content generation.

Workflow

When /review is invoked, execute the following steps in order.

---

Step 1: Check for today's review file

Run:

cat data/review/$(date +%Y-%m-%d).md 2>/dev/null

If the file exists, display a brief summary to Robin:

"Today's pulse found {N} ideas. Here's the overview:"

Then show the file content (or a condensed version with titles and sources).

If no file exists, tell Robin:

"No pulse has run today. Run /pulse first, or I can review ideas from previous days."

Continue to Step 2 regardless — there may be pending ideas in the DB from earlier pulse runs.

---

Step 2: Query pending ideas from the database

Run:

node -e "
  const db = require('better-sqlite3')('data/content.db');
  const ideas = db.prepare(\"SELECT id, title, summary, source_url, source_type, score, status, transcript FROM ideas WHERE status IN ('new', 'starred') ORDER BY score DESC\").all();
  console.log(JSON.stringify(ideas));
  db.close();
"

Parse the JSON output. If the array is empty, tell Robin:

"No pending ideas to review. All caught up! Run /pulse to discover new ideas."

Stop here if nothing to review.

---

Step 3: Present ideas and collect decisions

Initialize counters: kept=0, skipped=0, starred=0, remaining=total_count.

For each idea (iterate in order, score DESC):

Display this format:

--- Idea {N}/{total} ---

{title}

Source: {source_type} | {source_url}
Angle:  {content_angle_suggestion}

> {summary}

[Transcript available: yes/no]

Content angle suggestion logic (evaluate in order, first match wins):

• source_type === 'changelog' → "News reaction"
• title contains (case-insensitive) "tutorial", "how to", "guide", or "learn" → "Tutorial breakdown"
• title contains (case-insensitive) "wrong", "actually", "unpopular", "hot take", or "controversial" → "Hot take"
• title contains (case-insensitive) "vs", "compare", "comparison", or "better" → "Comparison"
• default → "Topic breakdown"

DO NOT show score or score breakdown. (Per D-09 — Robin doesn't need to see the math.)

After presenting the idea, use AskUserQuestion to ask:

"KEEP (k) — generate content | SKIP (s) — dismiss | STAR (st) — save for later | QUIT (q) — stop reviewing"

Accept these inputs case-insensitively:

• k or keep → status: kept
• s or skip → status: skipped
• st or star → status: starred
• q or quit → stop the loop, go to Step 5

---

Step 4: Update status in database (after each decision)

After Robin makes a decision for an idea, immediately run:

node scripts/review-update.js --id <actual_idea_id> --status <actual_status>

Where <actual_idea_id> is the real UUID from the query results and <actual_status> is kept, skipped, or starred.

CRITICAL: Never use inline node -e with placeholder strings like '{id}' or '{status}' for DB updates — always use scripts/review-update.js with the actual values interpolated into the CLI arguments. The script uses a parameterized UPDATE query.

Update counters after each decision:

• kept++ if kept
• skipped++ if skipped
• starred++ if starred
• remaining-- for each idea reviewed

If Robin typed quit, stop iterating and go directly to Step 5.

Step 4a: Capture visual approach when Robin KEEPs an idea (D-02)

When Robin selects k or keep, immediately after running review-update.js, ask:

"Visual approach for this topic?"

• photo — Use photos from the catalog with text overlays (default if catalog has photos)
• ai — Generate slide images with AI (Gemini)

Accept photo or ai (case-insensitive). Default to photo if Robin just presses Enter.

First, ensure the visual_approach column exists (idempotent — safe to run every time):

node -e "
  const Database = require('better-sqlite3');
  const db = new Database('data/content.db');
  try { db.exec('ALTER TABLE ideas ADD COLUMN visual_approach TEXT'); } catch(e) { if (!e.message.includes('duplicate column')) throw e; }
  db.close();
"

Then update the idea with Robin's visual approach choice:

node scripts/review-update.js --id <actual_idea_id> --status kept --visual-approach <photo|ai>

This stores Robin's visual approach in the database so the content-orchestrator agent can read it at content generation time (via idea.visual_approach). This is D-02: visual approach chosen at KEEP time, not at generation time.

---

Step 5: Show review summary

After all ideas are reviewed (or Robin quits), display:

Review complete!
- Kept:      {kept} ideas (ready for content generation)
- Skipped:   {skipped} ideas
- Starred:   {starred} ideas (will appear in next review)
- Remaining: {remaining} ideas (not yet reviewed)

If kept > 0, add:

"Run /generate-content on any kept idea to produce TikTok EN, TikTok DE, Instagram, and LinkedIn drafts."

---

Important Behavioral Rules

• D-08: Decisions happen ONLY via this interactive command. Never instruct Robin to edit the markdown file.
• D-09: Show title + summary, source + link, and content angle. Do NOT show the numeric score.
• Previously starred ideas appear alongside new ideas (status IN ('new', 'starred')), sorted by score DESC. No separate view needed.
• Transcript content is not shown inline during card review. If Robin asks about a specific idea's transcript, run:

  node -e "const db=require('better-sqlite3')('data/content.db'); const r=db.prepare('SELECT transcript FROM ideas WHERE id=?').get('<id>'); console.log(r && r.transcript || 'No transcript available.'); db.close();"
  

  and display the result.
• Quit at any time: Robin can type q or quit to exit the loop immediately and see the summary.

---

Schedule

Time	Command	Phase	Status
6:00 AM	/pulse	Phase 2	Active
Manual	/review	Phase 2	Active
Manual	/generate-content	Phase 3	Active
6:00 PM	perf-check	Phase 4	Not yet created

Registration commands (per D-07, using Claude Code /schedule):

/schedule "0 6 * * *" /pulse
/schedule "0 18 * * *" perf-check

Note: /review and content generation skills are manual — no cron registration needed.

---

Dependencies

• data/content.db — ideas table populated by /pulse (status='new' records)
• scripts/review-update.js — CLI script for safe DB status updates
• Ideas must exist with status='new' or 'starred' — requires /pulse to have run at least once