claude-social-pipeline

Capture insights while coding. Draft tweets without leaving the terminal.

Genuine insights happen while you're deep in code. By the time you open Twitter, the moment's gone and you're staring at a blank box. This captures them in-context and drafts content without breaking flow.

---

claude plugins install nardovibecoding/claude-social-pipeline

---

What It Does

The best content comes from real work. This plugin makes it frictionless to capture insights as they happen, humanize drafts before publishing, and post from the terminal — without breaking flow.

The pipeline:

coding session
     │
     ├── interesting thing happens
     │       └── content_capture("discovered X")
     │                   └── ~/.claude/content_drafts/running_log.md
     │
     ├── session getting long
     │       └── session_checkpoint("built Y")
     │                   └── ~/.claude/content_drafts/checkpoint_ts.md
     │
     ├── task complete
     │       └── post_task_check()
     │                   └── "content-worthy: use content_capture"
     │
     └── ready to post
             └── /tweet suggest
                         └── git log → privacy filter → draft
                                     └── humanizer → approve → X API

---

MCP Tools

Six tools Claude calls automatically during your session — no slash commands needed.

Tool	What it does
content_capture	Save a tweet-worthy moment mid-session. Appends to running_log.md with timestamp and context.
content_queue	Manage tweet drafts — add, list, get next, mark posted. Priority-sorted queue.
session_checkpoint	Snapshot session state (summary, decisions, files changed) before /clear or context limit.
post_task_check	After finishing a task — scan session for content-worthy material and improvement patterns.
set_reminder	Terminal alert with no background service: "30m", "2h", or "16:55" (HKT).
tweet_performance	Pull engagement stats for recent tweets — likes, retweets, replies, views, engagement score. Auto-captures best performer to content log.

---

Skills

/tweet — x-tweet

Draft, humanize, and post tweets. Runs every post through a multi-step pipeline before anything reaches the API.

Pipeline: git log → privacy filter → voice rules → content-humanizer → anti-pattern scan → manual approval → X API v2

Modes:

Command	What it does
/tweet [topic]	Draft from a topic, approve, post
/tweet suggest	Read git log, surface privacy-safe angles
/tweet hot	Search X for trending topics in your space
/tweet draft	Save to queue without posting
/tweet queue	View and pick from saved drafts
/tweet thread	Generate a 3–5 tweet thread

Never auto-posts. Every tweet requires explicit approval before the X API is called.

Env vars required:

X_API_KEY=your_key
X_API_SECRET=your_secret
X_ACCESS_TOKEN=your_token
X_ACCESS_TOKEN_SECRET=your_token_secret

---

content-humanizer

Detects and removes AI writing patterns. Not a word-swap tool — it rebuilds voice from the ground up using a two-pass approach.

Triggers: "sounds like AI", "make it human", "add personality", "too generic", "fix AI writing"

How it works:

1. humanizer_scorer.py runs a 12-point checklist: filler words, hedge phrases, em-dash overuse, passive constructions, hollow intensifiers, and more
2. Produces a score (0–100) and a list of specific tells
3. Rewriter addresses each tell with concrete changes — not paraphrasing

Before / After:

BEFORE (AI score: 31/100)
"It's worth noting that this approach leverages cutting-edge techniques
to significantly enhance performance. The implementation showcases a
robust solution that addresses the core challenges effectively."

AFTER (AI score: 89/100)
"This drops response time by 40%. The trick: cache the embedding lookup
instead of recomputing on every request. Obvious in hindsight."

Attribution: Content humanizer skill originally by Alireza Rezvani — MIT licensed. Scorer script and voice techniques added for this distribution.

---

Install

One command. Takes 30 seconds.

curl -fsSL https://raw.githubusercontent.com/nardovibecoding/claude-social-pipeline/main/install.sh | bash

Clones the repo, installs mcp package, registers MCP server in ~/.claude/settings.json. Restart Claude Code.

Manual install

git clone https://github.com/nardovibecoding/claude-social-pipeline.git
pip install mcp

Add to ~/.claude/settings.json:

{
  "mcpServers": {
    "content-pipeline": {
      "command": "python3",
      "args": ["/path/to/claude-social-pipeline/mcp/server.py"]
    }
  }
}

Copy skills:

cp -r claude-social-pipeline/skills/x-tweet ~/.claude/skills/
cp -r claude-social-pipeline/skills/content-humanizer ~/.claude/skills/

---

Configuration

No config required for basic usage (content capture, checkpoints, humanizer).

For x-tweet posting, set in .env:

X_API_KEY=your_key
X_API_SECRET=your_secret
X_ACCESS_TOKEN=your_token
X_ACCESS_TOKEN_SECRET=your_token_secret

For Telegram post alerts (optional):

TELEGRAM_BOT_TOKEN=your_bot_token
TELEGRAM_CHAT_ID=your_chat_id

---

Project Structure

claude-social-pipeline/
├── mcp/
│   ├── server.py              # MCP server — 6 content tools
│   ├── lib.py                 # Pure Python content functions
│   ├── patterns.py            # post_task_check patterns
│   └── pyproject.toml
├── skills/
│   ├── x-tweet/
│   │   ├── SKILL.md
│   │   ├── scripts/
│   │   │   ├── post_tweet.py          # X API v2 posting
│   │   │   └── tweet_stats.py         # Performance metrics
│   │   └── references/
│   │       ├── voice-rules.md
│   │       ├── templates.md
│   │       ├── anti-patterns.md
│   │       ├── hashtag-strategy.md
│   │       ├── content-calendar.md
│   │       └── engagement-data.md
│   └── content-humanizer/
│       ├── SKILL.md
│       ├── scripts/
│       │   └── humanizer_scorer.py
│       └── references/
│           ├── ai-tells-checklist.md
│           └── voice-techniques.md
├── README.md
└── LICENSE

Content drafts are written to ~/.claude/content_drafts/:

~/.claude/content_drafts/
├── running_log.md       # All content_capture() moments
├── queue.md             # Tweet draft queue (priority-sorted)
└── checkpoint_*.md      # Session checkpoints

---

Star History

---

License

AGPL-3.0 — see LICENSE.

content-humanizer skill includes work originally by Alireza Rezvani (MIT). See skills/content-humanizer/SKILL.md for full attribution.