| name | video-analyze |
| description | OpenClaw-aware leverage analysis of video transcripts. Produces structured JSON with leverage points, easy wins, cost analysis, discussion seeds, and auto-proposed ideas. |
| version | 1.0.0 |
| tags | ["video","analyze","leverage","transcript","ideas","strategy","openclaw"] |
video-analyze — Video Leverage Analysis
Purpose
Analyze a video transcript for OpenClaw leverage opportunities. This is not generic summarization — it is system-aware strategic analysis that maps video content to our 18-agent architecture.
When Used
- Called by
video-discuss.py host script with a full transcript (preferred path)
- Can also be triggered by agentToAgent from Relay, Eoin, Research, or Captain
- Input: video title, channel, date, trust level, transcript text
Self-Ingest Fallback
If you receive a YouTube URL or video ID WITHOUT a transcript, you MUST fetch it yourself before analyzing:
- Check DB first:
exec python3 -c "import sqlite3; db=sqlite3.connect('/root/.openclaw/transcripts.db'); r=db.execute('SELECT title, transcript FROM videos WHERE video_id=?', ('<VIDEO_ID>',)).fetchone(); print(r[0] if r else 'NOT_FOUND')"
- If in DB: read the transcript from the query result and proceed to analysis
- If NOT in DB: ingest it:
exec python3 /root/.openclaw/scripts/youtube-ingest.py <url>
- Then re-read from DB and proceed
This fallback ensures you can always analyze a video even if the caller skips the host script. The ingestion uses Gemini Flash API (~$0.01-0.05).
Process
-
Search Chartroom first — use chart_search to check current system state for any topics the video covers. This grounds your analysis in reality.
-
Analyze for leverage — map video insights to our architecture:
- Which of our 18 agents could benefit?
- Which of our 27 crons could be improved?
- Does this affect our engine routing (Helm, Codex, Flash)?
- Are there new tool/skill ideas?
- Cost implications (we run on ~$23-25/mo)?
-
Score and propose ideas — every actionable insight becomes a scored idea:
- Impact (1-5): how much value does this add?
- Effort (1-5, 5=easiest): how hard to build?
- Urgency (1-5): time-sensitive or evergreen?
- Minimum score product of 27 to include
-
Generate discussion seeds — 3-5 questions that help Robert think about applying insights. Frame as decisions, not information dumps.
Output Format
Return ONLY valid JSON:
{
"summary": "2-3 sentence summary",
"key_insights": ["insight 1", "insight 2", "..."],
"leverage_points": [
{"point": "description", "agent": "agent-id", "difficulty": "easy|medium|hard", "cost": "free|low|medium"}
],
"easy_wins": ["free change 1", "free change 2"],
"cost_effective": [
{"idea": "description", "estimated_cost": "$X/mo", "roi_reasoning"
Trust-Weighted Analysis
- Trust >= 0.7: Standard confident framing. "We should..." / "This means..."
- Trust < 0.7: Cautious framing. "Worth investigating..." / "If this holds up..."
- Trust < 0.3: Skeptical framing. "Claims to be..." / "Would need verification..."
Low-trust ideas get flagged in their description text so Robert can see the caveat in the pipeline.
Constraints
- UNMANNED ONLY: every idea must work without Robert's active involvement
- Score honestly — don't inflate to get ideas into the pipeline
- Reference Chartroom findings in leverage points where relevant
- Keep summary under 3 sentences
- 3-5 insights, not 10
- Return ONLY JSON — no markdown wrapper, no explanation text
Intent: Resourceful [I07], Informed [I06]. Purpose: Map external intelligence to internal leverage.
