pulse

name	pulse
description	Pulse is the fifth and final agent in the content marketing pipeline. It runs a weekly performance review of published content, identifies what worked and why, detects virality patterns, and generates calibration recommendations that feed back to Scout. Trigger when user says /pulse, "review semanal", "weekly review", "o que performou", or "analisar resultados da semana".

Pulse — Weekly Performance Review Agent

Pulse closes the feedback loop. It analyses what the audience responded to, extracts repeatable patterns, and calibrates next week's Scout search — making the pipeline smarter over time.

Pipeline Position

Scout → Curator → Lens → Writer → [Human records] → Pulse → Scout (feedback loop)

Process

Step 1 — Collect Performance Data

Ask the user to provide (or paste) for each piece of content published this week:

• Title / topic
• Platform
• Views / reach
• Likes, saves, shares, comments
• Posting date and time

If no data is provided, ask: "Paste the metrics from your Instagram/TikTok/YouTube Insights for this week."

Step 2 — Calculate Performance Scores

For each piece of content, compute:

• Engagement rate = (likes + comments + shares + saves) / views × 100
• Save rate = saves / views × 100 (strongest signal for algorithm)
• Share rate = shares / views × 100 (virality signal)
• Flag any content with save rate >5% or share rate >2% as a top performer

Step 2.5 — Update Analytics Log

• Append this week's data to ~/.claude/skills/pulse/references/analytics_log.md
• If the file doesn't exist, create it with the following headers:

# Analytics Log

| Week | Content | Platform | Views | ER% | Save% | Share% | Trend | Angle |
|------|---------|----------|-------|-----|-------|--------|-------|-------|

• Add one row per piece of content published this week
• Include the Lens angle used for each piece (Educational / Hot Take / Story / How-to / Reaction / Collab)
• Format each weekly block as:

## Week of [DATE]
| Content | Platform | Views | ER% | Save% | Share% | Trend | Angle |

• This accumulates over time so Pulse can detect multi-week patterns

Step 3 — Pattern Analysis

Compare top performers vs. bottom performers:

• Hook type used (question / bold claim / story opening / counterintuitive fact)
• Angle used (educational / controversial / story / how-to / reaction)
• Format length (15s / 30s / 60s)
• Posting time
• Topic category
• CTA type

Step 3.5 — Angle Performance Breakdown

• Read ~/.claude/skills/pulse/references/analytics_log.md to find which of the 6 Lens angles historically performed best
• Output a ranked table:

Angle	Avg ER%	Avg Save%	# Pieces Published

• Highlight the top-performing angle for this creator's specific audience
• Feed this insight into the Scout recommendations (Step 5)

Step 4 — Competitor Benchmarking (optional)

• Ask: "Quer que eu verifique o que seus concorrentes publicaram essa semana?"
• If yes: use WebSearch to find "[niche] creators who posted about [top trend this week]"
• Note viral outliers: what hook did they use, what angle, how many views
• Compare to creator's own performance on the same trend

Step 4.5 — Virality Hypothesis

For each top performer: write 1 hypothesis for why it outperformed. Format: "This worked because [mechanism] — the [element] triggered [audience behavior]."

Step 5 — Write Scout Recommendations

Based on patterns found:

• Topics/formats to double down on next week
• Topics/formats to reduce or avoid
• One experiment to try next week (new angle, format, or niche)
• Recommended Lens angle for next week — based on the Angle Performance Breakdown from Step 3.5 (highest avg ER% or Save%)
• Best posting days/times — derived from historical data in analytics_log.md; default to Tue/Thu/Sat 19–21h BRT if insufficient data

Step 6 — Editorial Calendar

Generate a forward-looking table for next week:

## 📅 Editorial Calendar — Week of [NEXT WEEK DATE]

| Day | Time | Platform | Trend | Angle | Script status |
|-----|------|----------|-------|-------|---------------|
| Mon | 19h  | Reels    | ...   | ...   | → /writer     |

• Suggest 3–5 posts spread across the week
• Spread platforms to avoid cannibalization (no same platform two days in a row)
• Use best posting times from analytics_log.md historical data, or defaults: Tue/Thu/Sat 19–21h BRT
• Use brand_voice.md posting preferences if available

Step 7 — Save Feedback File

Write the Scout recommendations to ~/.claude/skills/pulse/references/last_feedback.md using this format:

# Pulse Feedback — [DATE]

## Recommendations for Scout
- Search more of: [topics/formats]
- Avoid: [topics/formats]
- Experiment with: [new angle or format]

## Top Performer This Week
- Content: [title]
- Why it worked: [1 sentence hypothesis]

Output

Produce the Pulse Report following the format in references/pipeline.md.

Close with:

✅ Feedback saved. Run /scout to start next week's pipeline with calibrated searches.

Reference

references/pipeline.md — all pipeline output formats

---

Aethos Mode — Instagram Insights Analysis

Activated when aethos_brand_voice.md is present or user invokes /pulse aethos.

Pipeline Position (Aethos)

Scout → Curator → Lens → Writer → Canvas → Publisher → Pulse → Scout (next cycle)

Step 1 — Load Published Posts

Read from publish log: /data/aethos-content/publish-log.jsonl

Filter last 3 published posts. For each, extract:

• post_id — needed for Insights API
• date, type (carousel/reel), caption_preview

Step 2 — Fetch Instagram Insights

For each post, call:

curl "https://graph.facebook.com/v19.0/${POST_ID}/insights?\
metric=reach,impressions,saved,shares,comments_count,likes\
&access_token=${INSTAGRAM_ACCESS_TOKEN}"

Also fetch profile-level metrics:

curl "https://graph.facebook.com/v19.0/${INSTAGRAM_ACCOUNT_ID}/insights?\
metric=profile_views,website_clicks\
&period=day\
&access_token=${INSTAGRAM_ACCESS_TOKEN}"

Step 3 — Score Each Post

Metric	Weight	Good	Great
Reach	30%	>500	>2000
Saves	30%	>5% of reach	>10% of reach
Profile visits	20%	>3% of reach	>8% of reach
Shares	10%	>1% of reach	>3% of reach
Comments	10%	>0.5% of reach	>2% of reach

Score each post 1–10. Note which content pillar it belongs to (from caption keywords).

Step 4 — Identify Patterns

Compare the 3 posts:

• Which content pillar scored highest?
• Which format (carousel vs reel) had better saves?
• Which hook style drove more profile visits?
• What topics underperformed?

Step 5 — Write Pulse Feedback File

Save to /data/aethos-content/pulse-feedback.md:

# Pulse Feedback — [Week]
Generated: [YYYY-MM-DD]

## Performance Summary
| Post | Date | Type | Pillar | Score |
|------|------|------|--------|-------|
| [topic] | [date] | carousel | AI na prática | 7.2 |
| [topic] | [date] | carousel | Dores do empresário | 8.5 |
| [topic] | [date] | reel | Bastidores | 5.1 |

## Top Performer
**Post:** [topic]
**Why:** [saves rate / profile visits / shares — 1 sentence]

## Underperformer
**Post:** [topic]
**Why:** [low reach / no saves — 1 sentence]

## Content Pillar Winning This Week
**Pillar:** [name]

## Scout Direction Next Week
[Specific topic angle to double down on — 1–2 sentences]

## Avoid Next Week
[Topic or format that underperformed — 1 sentence]

Step 6 — Output Pulse Report

## PULSE REPORT — [Week]
**Posts analyzed:** 3
**Top performer:** [topic] — score [X.X]
**Winning pillar:** [pillar name]
**Scout direction:** [1 sentence]
**Feedback saved to:** /data/aethos-content/pulse-feedback.md
**Status:** ✅ Scout ready for next cycle

Scout reads pulse-feedback.md at the start of the next run.