Blog Writer Skill
Autonomous blog writer that researches your industry and writes SEO + AEO optimized drafts. Designed to run unattended via /monitor cron.
This skill does NOT modify code. It reads product context, researches the industry, writes content, saves to the database, and emails the admin.
Prerequisites
- Blog infrastructure set up (
/blogcommand completed) TAVILY_API_KEYin env (free at tavily.com — 1,000 searches/month)ADMIN_EMAILinwebapp/.envRESEND_API_KEYinwebapp/.envSUPABASE_SECRET_KEYinwebapp/.env- Jina MCP Reader available (for full content extraction)
Scripts
All scripts read webapp/.env automatically. Run from project root.
| Script | What it does | Usage |
|---|---|---|
library/skills/search/scripts/search.sh |
Tavily web search | search.sh '{"query": "...", "max_results": 10}' |
library/skills/blog-writer/scripts/get-recent-posts.sh |
Fetch last N posts from DB | get-recent-posts.sh 20 → JSON array |
library/skills/blog-writer/scripts/save-post.sh |
Insert draft into blog_posts | save-post.sh '{"slug":...,"title":...}' |
library/skills/blog-writer/scripts/notify-admin.sh |
Email admin about new draft | notify-admin.sh '{"title":...,"slug":...}' |
Phase 1: Load Product Context
Read these files to understand the product, personas, and niche:
docs/project/product-overview.md → product name, category, pricing, value prop
docs/project/users-and-journeys.md → target personas, their struggles, goals
docs/project/feature-map.md → what the product does (for CTA relevance)
docs/project/brand-guidelines.md → voice, tone, author info
Extract and hold in memory:
- Product name + category + value prop
- Personas — who they are, their pain points, what they search for
- Industry/niche — the space this product operates in
- Brand voice — how to write (tone, formality, personality)
- Author info — name, role, avatar URL for byline
Phase 2: Check Existing Posts (Dedup)
Query the last 20 blog posts to avoid duplicating topics.
library/skills/blog-writer/scripts/get-recent-posts.sh 20
Returns JSON array: [{title, description, slug, category, created_at}]
The script reads webapp/.env automatically for Supabase credentials. Uses service role key to see drafts too.
Store the list of existing titles + descriptions. You MUST NOT write about a topic already covered.
Phase 3: Research — Find the Perfect Topic
This is the most critical phase. You're finding what's valuable to the personas, not what's easy to write.
Step 3a: Industry Research via Tavily
Run 3-5 Tavily searches targeting different angles:
# 1. Persona struggles + solutions
library/skills/search/scripts/search.sh '{"query": "[persona] struggles with [pain point] [current year]", "max_results": 10, "search_depth": "advanced", "time_range": "month"}'
# 2. Industry trends + news
library/skills/search/scripts/search.sh '{"query": "[industry/niche] trends [current year]", "max_results": 10, "search_depth": "advanced", "time_range": "month"}'
# 3. High-converting content in the niche
library/skills/search/scripts/search.sh '{"query": "best [product category] guide [current year]", "max_results": 10, "search_depth": "advanced"}'
# 4. Questions people are asking
library/skills/search/scripts/search.sh '{"query": "[niche] frequently asked questions how to", "max_results": 10, "search_depth": "advanced"}'
# 5. Competitor content gaps
library/skills/search/scripts/search.sh '{"query": "[product category] vs alternatives comparison", "max_results": 10, "search_depth": "advanced"}'
Replace bracketed terms with actual product context from Phase 1.
Step 3b: Analyze Results & Pick Topic
From the search results, identify:
- Topics with high search intent — questions people actually ask
- Gaps — topics NOT covered by existing posts (Phase 2 dedup)
- Conversion potential — topics where the product is a natural solution
- AEO opportunity — topics where a well-structured answer could get cited by AI
Topic selection criteria (ranked):
- Directly addresses a persona pain point
- Has clear search volume (appears in multiple results)
- Product is a natural, non-forced solution
- Not already covered in existing posts
- Can include data/stats (GEO boost)
- Suitable for FAQ or How-To schema (AEO boost)
Step 3c: Deep Research via Jina
For the chosen topic, extract full content from the 2-3 best sources:
Use Jina MCP Reader to get markdown content from URLs found in Tavily results. This gives you:
- Current statistics and data points to cite
- Expert quotes to reference
- Detailed information for comprehensive coverage
Cost efficiency: Only use Jina for 2-3 URLs max per post. Tavily snippets are enough for topic selection; Jina is for depth.
Phase 4: Write the Post
Content Structure
Every post MUST follow this structure for maximum SEO + AEO performance:
# [Title — matches target query naturally]
<Callout>
[40-60 word extractable summary answering the core question. This is the featured snippet candidate.]
</Callout>
<LastUpdated date="[today's date]" />
## [Section matching "What is X?" or core definition]
[Definition block — concise, self-contained, citable]
## [Problem/Pain section — why this matters]
[Evidence sandwich: claim → data points with sources → actionable insight]
<Stat value="[number]" source="[org/study]" date="[year]">
[Context for the statistic]
</Stat>
## [Solution/How-to section]
<Steps>
1. **[Step name]**: [Clear action in 1-2 sentences]
2. **[Step name]**: [Clear action in 1-2 sentences]
3. **[Step name]**: [Clear action in 1-2 sentences]
</Steps>
## [Comparison or evaluation section — if applicable]
<ComparisonTable>
| Feature | Option A | Option B |
|---------|----------|----------|
| ... | ... | ... |
**Bottom line**: [1-2 sentence recommendation]
</ComparisonTable>
<ExpertQuote author="[Name]" role="[Title]" org="[Company]">
[Relevant quote from industry expert found in research]
</ExpertQuote>
## [Practical tips / Best practices]
[Listicle or numbered list with actionable advice]
<CTA>
[Natural product mention — how {product name} helps with this specific problem. NOT salesy. Helpful.]
</CTA>
## Frequently Asked Questions
<FAQ>
### [Question phrased as users search]?
[Direct answer first sentence. Supporting context 2-3 sentences.]
### [Question phrased as users search]?
[Direct answer first sentence. Supporting context 2-3 sentences.]
### [Question phrased as users search]?
[Direct answer first sentence. Supporting context 2-3 sentences.]
</FAQ>
Writing Rules
- Voice: Match brand guidelines exactly. No generic AI tone.
- Length: 1,500-2,500 words. Comprehensive but not padded.
- Stats: Every claim backed by data with source + date. Use stats from Jina research.
- AEO blocks: Every post needs at least: 1 Callout, 1 FAQ (3+ questions), 1 Stat. Add Steps/ComparisonTable/ExpertQuote where natural.
- Target query: Pick ONE primary query the post targets. This becomes
target_queryin DB. - FAQ data: Extract FAQ questions + answers as structured data for FAQPage schema.
- Schema types: Default
['Article']. Add'FAQPage'if FAQ exists. Add'HowTo'if Steps exists. - CTA: ONE natural product mention per post. Must solve the reader's problem, not pitch features.
- No fluff: Every paragraph earns its place. Cut anything that doesn't inform, convince, or convert.
- Humanize: Write like a knowledgeable human, not a content mill. Include opinions, specific recommendations, and real examples.
Phase 5: Generate Metadata
For each post, generate:
| Field | How to generate |
|---|---|
slug |
Kebab-case from title, max 60 chars |
title |
Post title (60-70 chars for SERP) |
description |
Meta description (150-160 chars, includes target query) |
content |
Full MDX content from Phase 4 |
category |
Infer from topic (e.g., "Industry Trends", "How-To", "Guide", "Comparison") |
tags |
3-5 relevant tags |
status |
'draft' (always — admin publishes) |
author_name |
From brand guidelines |
author_role |
From brand guidelines |
author_avatar |
From brand guidelines |
seo_title |
Optimized <title> override if different from title |
seo_description |
Override meta description if needed |
target_query |
Primary AEO target query |
schema_types |
['Article', 'FAQPage'] minimum (add 'HowTo' if Steps used) |
faq_data |
[{question, answer}] extracted from FAQ section |
Phase 6: Save to Database
Save the post using the provided script. Build the JSON payload from Phase 5 metadata:
library/skills/blog-writer/scripts/save-post.sh '{
"slug": "your-generated-slug",
"title": "Your Post Title",
"description": "Meta description 150-160 chars",
"content": "Full MDX content here...",
"category": "Guide",
"tags": ["tag1", "tag2", "tag3"],
"author_name": "Author Name",
"author_role": "Author Role",
"author_avatar": "https://...",
"target_query": "how to solve X",
"schema_types": ["Article", "FAQPage"],
"faq_data": [{"question": "What is X?", "answer": "X is..."}]
}'
The script:
- Reads
webapp/.envfor Supabase credentials automatically - Requires
SUPABASE_SECRET_KEY(bypasses RLS to insert) - Forces
status: "draft"(safety — never publishes directly) - Returns the created post as JSON
Phase 7: Email Notification
Notify the admin that a new draft is ready:
library/skills/blog-writer/scripts/notify-admin.sh '{
"title": "Your Post Title",
"slug": "your-generated-slug",
"category": "Guide",
"target_query": "how to solve X",
"word_count": 1800,
"description": "Meta description"
}'
The script:
- Reads
webapp/.envforRESEND_API_KEY,ADMIN_EMAIL,NEXT_PUBLIC_SITE_URL - Sends a styled HTML email with post details, preview link, and dashboard link
- Gracefully skips if
RESEND_API_KEYorADMIN_EMAILnot set (no crash)
Topic Ideas by Category
When selecting topics, rotate between these high-performing categories:
| Category | Example Format | AEO Potential |
|---|---|---|
| How-To Guide | "How to [solve persona pain point]" | HowTo schema, Steps component |
| Industry Trends | "[Year] [Industry] Trends: What's Changing" | Stats, ExpertQuotes |
| Comparison | "[Solution A] vs [Solution B]: Which is Right for You?" | ComparisonTable, FAQ |
| Best Practices | "[Number] [Industry] Best Practices for [Goal]" | Listicle, Steps |
| Problem-Solution | "Why [Persona] Struggle with [X] (And How to Fix It)" | FAQ, Callout, CTA |
| Data/Research | "[Industry] Statistics: [Number] Facts You Need to Know" | Stats, Evidence Sandwich |
| Beginner Guide | "Complete Guide to [Topic] for [Persona]" | Definition, Steps, FAQ |
| Tools/Resources | "Top [Number] [Tools/Resources] for [Goal] in [Year]" | ComparisonTable, listicle |
Scheduled drafting
Create a recurring task only when the user explicitly asks. Prefer a Codex app automation whose prompt invokes the enabled blog-writing skill and saves a draft. In a CLI-only environment, codex exec -C /absolute/project/path "Use the enabled blog-writer skill to create one draft" inherits the user's configured model and sandbox. Sending email or publishing remains a separate explicit action.
What This Skill Does NOT Do
- Modify any code files
- Publish posts (always saves as draft)
- Manage existing posts (that's the admin UI)
- Set up blog infrastructure (that's the
/blogcommand) - Handle images or media (cover images are OG-generated)