post: Generate Social Media Content

This skill creates authentic X (Twitter) posts about development work using a 4-step quality pipeline.

Usage

bash

# Analyze recent conversation for post-worthy content
/post

# Generate post about specific topic
/post "Created 5 custom skills for Kotlin workflow"

# Auto-analyze and suggest topics
/post auto

The 4-Step Pipeline

Follow the pipeline defined in .claude/rules/Social.md:

Step 1: Fact Collection

CRITICAL: Without concrete facts, STOP and ask the user.

Scan the recent conversation (last 50-100 messages) for:

•✅ Numbers (metrics, LOC, time, percentages)
•✅ Tools (technologies, libraries, versions)
•✅ Results (outcomes, changes, improvements)
•✅ User's actual phrases/quotes
•✅ Context (problem solved, why it matters)

If insufficient facts:

code

I need more concrete facts to write an authentic post. Can you provide:

1. Specific numbers/metrics?
2. Tools/technologies used?
3. Measurable results/outcomes?
4. What problem this solved?

Without facts, the post will be generic AI-slop.

Output from this step:

code

Facts collected:
- [fact 1 with source]
- [fact 2 with source]
- [fact 3 with source]
- User quote: "[exact quote]"

Step 2: Writer Agent

Use Task tool with general-purpose agent to write the post.

Prompt for writer agent:

code

Write a post for X (Twitter) in English about this development work.

FACTS (use ONLY these, don't make up anything):
[paste facts from Step 1]

VOICE GUIDELINES:
- First person (I/we)
- Short sentences
- Lead with most interesting fact
- Specific numbers and tools
- Show don't tell (facts over adjectives)
- Casual, conversational
- Honest about failures
- No marketing language
- Max 280 characters (single tweet) or thread if needed

BANNED WORDS: See .claude/rules/Social.md Slop Gate section

Main rule: If it's not in the FACTS, don't write it.

Output format:
[post text]

Get the draft from the writer agent.

Step 3: Slop Gate Validation

Critical check against banned phrases and AI patterns.

Load the slop gate rules from .claude/rules/Social.md:

•Check against 67 banned phrases
•Check against Wikipedia AI patterns
•Check for generic statements that could apply to anything
•Check for marketing/influencer language

For each violation found:

code

❌ SLOP DETECTED: "[phrase]"
Line: "[sentence containing it]"
Pattern: [banned phrase / AI pattern / generic puffery]

If violations found:

•Count total violations
•If > 3: Send back to writer agent with feedback
•If 1-3: Mark as warnings, let user decide
•If 0: ✅ PASSED

Slop gate output:

code

Slop Gate Results:
✅ PASSED - No AI-slop detected
or
⚠️ WARNINGS - 2 minor issues:
  - Line 1: "However" (transition word overuse)
  - Line 3: Possible generic statement
or
❌ FAILED - 5 violations found:
  - [list violations]
  Sending back to writer for revision...

Step 4: Human Review

Present the final draft with all context:

markdown

## 📝 Draft Post Ready for Review

**Post:**
[final post text - copy-ready]

**Stats:**
- Characters: X/280
- Thread: [Yes (N tweets) / No]
- Reading time: ~X seconds

**Facts Used:**
[bulleted list of facts from conversation]

**Slop Gate:** [✅ PASSED / ⚠️ WARNINGS / ❌ FAILED]
[If warnings/failed: list issues]

**Source conversation:**
- Based on messages from [time range]
- Key topics: [list topics]

---

**Your options:**
1. ✅ **Approve** - Ready to post
2. ✏️ **Edit** - Make changes (I'll learn from them)
3. ❌ **Reject** - Not posting this (tell me why)
4. 🔄 **Revise** - Give me specific feedback to regenerate

What would you like to do?

Handle user response:

•Approve → Done
•Edit → Log changes for learning
•Reject → Log reason, learn from it
•Revise → Take feedback, go back to Step 2

Examples

Example 1: Feature Development Post

User invokes: /post

Step 1 - Facts collected:

code

- Created 5 custom Claude Code skills
- Skills: ktlint-fix, full-check, new-feature, compose-screen, spring-endpoint
- Each skill took ~15-20 minutes to create
- Enabled ktlint trailing comma rules
- Added named arguments guideline (3+ params)
- Project: Spring Boot + Compose Multiplatform
- Follows 200-line iteration rule
- User removed spring-endpoint (not competent to review)

Step 2 - Writer output:

code

Built 5 custom Claude Code skills for my Kotlin workflow today:

/ktlint-fix - auto-format + verify
/full-check - run all CI checks locally
/new-feature - scaffolds iteration plan (200 LOC max)
/compose-screen - ViewModel + UI state boilerplate

Each skill took ~15min to write. Now my entire CI pipeline runs in one command.

Step 3 - Slop gate:

code

✅ PASSED
- No banned phrases detected
- No AI patterns found
- Concrete facts present
- Authentic voice maintained

Step 4 - Human review:

code

## 📝 Draft Post Ready for Review

**Post:**
Built 5 custom Claude Code skills for my Kotlin workflow today:

/ktlint-fix - auto-format + verify
/full-check - run all CI checks locally
/new-feature - scaffolds iteration plan (200 LOC max)
/compose-screen - ViewModel + UI state boilerplate

Each skill took ~15min to write. Now my entire CI pipeline runs in one command.

**Stats:**
- Characters: 274/280
- Thread: No
- Reading time: ~12 seconds

**Facts Used:**
- 5 custom skills created
- Specific skill names and purposes
- 200 LOC iteration constraint
- Time metric: 15 min per skill
- Benefit: CI pipeline in one command

**Slop Gate:** ✅ PASSED

**Your options:**
1. ✅ Approve - Ready to post
2. ✏️ Edit - Make changes
3. ❌ Reject - Not posting
4. 🔄 Revise - Give feedback

What would you like to do?

Example 2: Technical Learning Post

User invokes: /post "Learned about ktlint trailing comma rules"

Facts needed - ASK USER:

code

I need more concrete facts about the ktlint trailing comma rules. Can you provide:

1. What specific rules did you enable?
2. Why did you enable them?
3. What problem does this solve?
4. Any metrics (files affected, issues found, etc.)?

Without these facts, I can only write generic AI-slop.

Error Handling

Insufficient facts:

•STOP and ask user for specific details
•List what facts are missing
•Explain why they're needed

Writer agent fails:

•Log the error
•Ask user if they want to try again or provide different facts
•Don't present a bad draft

Slop gate fails (>5 violations):

•Don't present to user
•Send back to writer with violations list
•Max 3 revision attempts, then ask user for help

User rejects repeatedly:

•Log patterns in rejections
•Ask for feedback: "What should I change about how I write posts?"
•Update approach based on feedback

Learning & Improvement

After each post (approved or rejected):

•
Log the outcome:
- •Facts used
- •Final text (if approved)
- •User edits (if edited)
- •Rejection reason (if rejected)
•
Identify patterns:
- •What facts lead to approved posts?
- •What language gets edited out?
- •What topics work best?
•
Adjust approach:
- •Update fact collection priorities
- •Refine voice guidelines
- •Add to slop gate if new AI patterns emerge
•
Ask for meta-feedback periodically:
- •"How's the post quality been?"
- •"Should I adjust the tone/length/style?"
- •"Any new topics you want to focus on?"

Integration with Development Workflow

This skill can be triggered:

•Manually by user: /post when they want to share something
•Auto-suggested after iterations (per Social.md rule)
•Scheduled (e.g., "Friday recap" posts)

The pipeline ensures quality while being fast enough not to interrupt the development flow.