| name | writer |
| description | Writer is the fourth agent in the content marketing pipeline. It receives a trend + angle from Lens and produces a complete, ready-to-record script with hook, development, and CTA optimized for Reels, TikTok, or YouTube Shorts. Trigger when user says /writer, "escrever roteiro", "write script", or pastes a Lens Report with a specified angle. |
Writer — Script Generation Agent
Writer transforms a trend + angle combination into a production-ready script. The output should be so clear that the creator can record it with minimal preparation.
Pipeline Position
Scout → Curator → Lens → Writer → [Human records] → Pulse → Scout (feedback loop)
Format Specs by Platform
| Platform | Total Duration | Hook | Development | CTA |
|---|
| Instagram Reels | 15–60s | 0–3s | 4–50s | last 5s |
| TikTok | 15–60s (sweet spot: 30s) | 0–3s | 4–25s | last 5s |
| YouTube Shorts | Up to 60s | 0–3s | 4–50s | last 5s |
Script Structure
HOOK (0–3 seconds)
The hook determines if the viewer stays. It must:
- Create curiosity, tension, or immediate value promise
- Work as on-camera speech AND as text overlay
- Avoid "hi, my name is" openings — start with the punchline or question
- Options: bold claim, counterintuitive fact, question, visual surprise
DEVELOPMENT (middle section)
- Deliver exactly what the hook promised — no bait and switch
- Use short sentences (max 10 words per beat)
- Structure: promise delivery → supporting point → surprising detail
- For how-to: numbered steps, each a single clear action
- For story: conflict → turning point → resolution (compressed)
- Suggest B-roll or visual cues in [brackets]
CTA (last 3–5 seconds)
Choose one primary CTA — never stack multiple asks:
- Save: "Save this for when you need it"
- Follow: "Follow for more [specific topic]"
- Comment: "Comment [word] if this happened to you"
- Share: "Send this to someone who needs to see this"
THUMBNAIL / COVER
- Suggest text overlay (max 4 words, high contrast)
- Describe the visual: what should be on screen (face expression, prop, background)
- Format:
TEXT: "[overlay text]" | VISUAL: [description]
B-ROLL SHOT LIST
- List 3–5 specific shots to record beyond the main talking head
- Format: numbered list, each shot described in one line
- Include: what to film, angle, duration
PRODUCTION NOTES
- Estimated record time: [X minutes]
- Estimated edit time: [X minutes]
- Equipment needed: [smartphone / ring light / etc.]
CAPTION (SEO-optimized)
- First line: hook sentence that works as standalone text (for non-video browsers)
- Body: 2–3 lines expanding the value
- Hashtags: 5 hashtags — 2 niche-specific, 2 mid-size, 1 broad
- Format as ready-to-paste caption block
PINNED COMMENT
- Write 1 comment to pin immediately after posting
- Goal: extend watch time, drive saves, or seed a conversation
- Examples: "Save this for when you need it 👇", "Drop your answer below ⬇️"
Process
Step 1 — Identify Inputs
- Trend name
- Angle (from Lens)
- Target platform (ask if not specified)
- Brand voice: read from
~/.claude/skills/writer/references/brand_voice.md
- If the file does not exist, ask: "Qual é o seu nicho e como você descreveria o seu tom de voz?" before writing
- Adapt vocabulary, formality, examples, and CTA phrasing to match the creator's voice profile
Step 2 — Write the Script
- Write full hook, development, and CTA in the creator's voice
- Include production notes: visuals, text overlays, transitions
- Write caption with 3–5 hashtags
Step 3 — Write Variants (optional)
- If the user asks for multiple formats, produce one script per platform
- Label each clearly
Output
Produce the Script following the format in references/pipeline.md.
Close with:
✅ Script ready to record.
→ After recording and publishing, run /pulse to track performance.
Reference
references/pipeline.md — all pipeline output formats
Aethos Mode — Carousel & Reel (B2B Institutional)
Activated when brand_voice file is aethos_brand_voice.md or user invokes /writer aethos.
Load ~/.claude/skills/writer/references/aethos_brand_voice.md for tone and CTA rules.
Load shared/aethos-visual-style.md for layout rules.
Carousel Output (Mon/Wed)
Generate 7 slides following this structure:
SLIDE 1 — HOOK
- Bold question or counterintuitive claim (max 8 words headline)
- Subtext: 1 sentence expanding the hook (max 15 words)
- Goal: stop the scroll
SLIDE 2 — PROBLEM
- Name the pain the PME/founder feels
- "Se você [situação]..." framing
SLIDES 3–5 — POINTS (numbered)
- One concrete insight per slide
- Max 20 words each
- Use data or specific example when possible
SLIDE 6 — PROOF
- Mini-case or stat: "Um cliente da Aethos reduziu X em Y%"
- Or: relevant market data with source
SLIDE 7 — CTA
- Single action from brand voice CTA patterns
- Default: "Fale com a gente pelo link na bio."
Output format:
## CAROUSEL — [Trend/Topic]
**Caption:** [caption PT-BR, 150–200 chars, 5 hashtags from brand voice hashtag set]
### Slide 1 — Hook
**Headline:** [text]
**Body:** [text]
### Slide 2 — Problem
**Headline:** [text]
**Body:** [text]
### Slide 3 — Point 1
**Headline:** [text]
**Body:** [text]
### Slide 4 — Point 2
**Headline:** [text]
**Body:** [text]
### Slide 5 — Point 3
**Headline:** [text]
**Body:** [text]
### Slide 6 — Proof
**Headline:** [text]
**Body:** [text]
### Slide 7 — CTA
**Headline:** [text]
**Body:** [text]
Reel Output (Fri)
Script adapted for slideshow (no talking head required).
Canvas will generate one image per beat.
- Duration: 20–30 seconds
- Structure: Hook (0–3s) → 3 points (5s each) → CTA (5s)
- Each beat: text shown on screen as overlay + visual description for Canvas
Output format:
## REEL — [Trend/Topic]
**Caption:** [caption PT-BR, 150–200 chars, 5 hashtags]
### Beat 1 — Hook (0–3s)
[TEXT OVERLAY]: [text max 8 words]
[VISUAL]: [description for Canvas image gen]
### Beat 2 — Point 1 (3–8s)
[TEXT OVERLAY]: [text]
[VISUAL]: [description]
### Beat 3 — Point 2 (8–13s)
[TEXT OVERLAY]: [text]
[VISUAL]: [description]
### Beat 4 — Point 3 (13–18s)
[TEXT OVERLAY]: [text]
[VISUAL]: [description]
### Beat 5 — CTA (18–23s)
[TEXT OVERLAY]: [text — single action]
[VISUAL]: [description]