Turn Claude Code into the ultimate X (Twitter) content engine. Every skill is grounded in how the Phoenix algorithm actually scores posts — not guesswork, but the documented signal weights from X's open-source ranking system.
sparX is a collection of Claude Code skills, agents, and deep reference material that transforms your terminal into a full X content studio: drafting, optimizing, scoring, scheduling, trend research, performance review, and visual content creation.
# Clone the repo
git clone https://github.com/epistates/sparX.git
cd sparX
# Open Claude Code
claude
# Draft your first post
/compose I just shipped a feature that cuts API latency by 73%
# Or optimize an existing draft
/optimize We released v2.0 with lots of improvements. Check it out at https://example.comThat's it. No API keys, no dependencies, no configuration. sparX works out of the box.
The X algorithm is not a black box. X open-sourced its recommendation system, and the 2026 Phoenix update is documented in detail. The scoring weights are known:
| Signal | Weight vs Like | What This Means |
|---|---|---|
| Author replies to comments | 75–150× | Reply to every comment in the first hour |
| Threaded conversations | 13–27× | Create content that sparks genuine discussion |
| Reposts / Quote tweets | ~20× | Make posts worth sharing with commentary |
| Bookmarks | ~10× | Create reference-worthy, save-able content |
| Dwell time | Very high | Write posts people actually spend time reading |
| Likes | 1× | The least valuable positive signal |
Most creators optimize for likes. The algorithm cares about conversations, bookmarks, and dwell time — signals that are 10–150× more valuable.
sparX embeds this knowledge into every skill, so you don't have to memorize the hierarchy. Just use the skills and the optimization is built in.
| Skill | Purpose | Example |
|---|---|---|
/compose |
Draft algorithm-optimized posts | /compose local LLM inference is now faster than cloud APIs |
/thread |
Create high-engagement threads | /thread 5 lessons from building a distributed cache |
/optimize |
Score and rewrite existing drafts | /optimize [paste your draft] |
/hooks |
Generate scroll-stopping openers | /hooks announcing a new developer tool |
/schedule |
Data-driven posting time recommendations | /schedule tech audience, EST timezone |
/research |
Discover trending topics and content opportunities | /research what's trending in the Rust ecosystem |
/review |
Analyze post-publish performance | /review 12K impressions, 47 replies, 89 bookmarks |
/strategy |
Plan weekly/monthly content strategy | /strategy dev tools niche, 500 followers, growth mode |
/repurpose |
Transform content across formats | /repurpose [blog post] → thread |
/media |
Visual content strategy and creation | /media what visual should I add to this benchmark post |
/post |
Publish directly to X via browser | /post [content] or compose then /post last |
/analyze |
Read real analytics from X via browser | /analyze recent or /analyze [post URL] |
/engage |
Reply to comments via browser (75–150× multiplier) | /engage [post URL] or /engage notifications |
/setup |
Install optional integrations | /setup remotion |
When you invoke a skill, Claude:
- Loads CLAUDE.md — essential Phoenix algorithm context (always present)
- Reads the skill instructions — step-by-step process for the specific task
- Progressively loads reference docs — deep algorithm data only when needed
- Reads supporting files — templates, patterns, examples, checklists
- Produces optimized output — scored, structured, ready to use
/compose → reads CLAUDE.md (algorithm weights)
→ reads compose/SKILL.md (composition process)
→ reads reference/scoring.md (signal details)
→ reads compose/templates.md (proven structures)
→ outputs: optimized post + score + alternatives
sparX/
├── .claude/
│ ├── skills/ ← 14 slash-command skills
│ │ ├── compose/ ← /compose: Draft optimized posts
│ │ │ ├── SKILL.md ← Composition process
│ │ │ ├── templates.md ← Proven post structures & hooks
│ │ │ └── examples.md ← High-performing post examples
│ │ ├── thread/ ← /thread: Create thread sequences
│ │ │ ├── SKILL.md ← Thread building process
│ │ │ └── structures.md ← 6 proven thread archetypes
│ │ ├── optimize/ ← /optimize: Score & rewrite drafts
│ │ │ ├── SKILL.md ← Scoring methodology
│ │ │ └── checklist.md ← Pre-publish optimization checklist
│ │ ├── hooks/ ← /hooks: Generate opening lines
│ │ │ ├── SKILL.md ← Hook generation process
│ │ │ └── patterns.md ← 8 hook pattern categories
│ │ ├── media/ ← /media: Visual content strategy
│ │ │ ├── SKILL.md ← Media tier system & creation guides
│ │ │ └── remotion-guide.md ← Remotion video integration
│ │ ├── post/ ← /post: Publish to X via browser
│ │ ├── analyze/ ← /analyze: Read analytics via browser
│ │ ├── engage/ ← /engage: Reply management via browser
│ │ ├── schedule/ ← /schedule: Posting time optimization
│ │ ├── research/ ← /research: Trend & topic discovery
│ │ ├── review/ ← /review: Performance analysis
│ │ ├── strategy/ ← /strategy: Content planning
│ │ ├── repurpose/ ← /repurpose: Format transformation
│ │ └── setup/ ← /setup: Integration installation
│ └── agents/
│ └── scorer.md ← Phoenix score simulation subagent
│
├── reference/ ← Deep algorithm knowledge base
│ ├── algorithm.md ← Complete Phoenix pipeline architecture
│ ├── scoring.md ← All signal weights & multipliers
│ ├── content-formats.md ← Format-specific optimization
│ ├── timing.md ← Data-driven posting time analysis
│ └── penalties.md ← What kills reach & how to avoid it
│
├── docs/ ← Comprehensive documentation
│ ├── getting-started.md ← First post in 5 minutes
│ ├── skills-guide.md ← Detailed usage for every skill
│ ├── algorithm-guide.md ← Creator-friendly algorithm explainer
│ ├── media-strategy.md ← Visual content strategy
│ ├── integrations.md ← Optional integration setup guide
│ └── playbooks.md ← End-to-end workflow walkthroughs
│
├── setup/
│ └── README.md ← Setup reference & file structure
├── CLAUDE.md ← Core algorithm context (always loaded)
└── README.md ← This file
Three skills turn sparX into a fully end-to-end system — draft, post, analyze, and engage without leaving Claude Code:
/compose my new feature cuts cold starts by 90%
↓ draft optimized, scored, ready
/post last
↓ opens X in Chrome, types the post, screenshots for your approval, publishes
↓ ...one hour later...
/engage [post URL]
↓ reads replies, prioritizes by value, drafts responses, posts with your approval
↓ ...next day...
/analyze [post URL]
↓ reads real metrics from X, runs Phoenix scoring analysis, suggests next content
| Skill | What It Does | Confirmation Required |
|---|---|---|
/post |
Types and publishes posts/threads in your browser | Yes — screenshots before every post |
/analyze |
Reads engagement metrics directly from X | No — read-only |
/engage |
Drafts and posts replies to comments | Yes — approval before each reply |
Requirements: claude-in-chrome extension installed. You must be logged into X in Chrome. The skills never handle authentication.
Every "For You" post goes through this pipeline:
Your post → shown to ~10-20% of followers → Phoenix AI scores engagement signals
↓
Strong signals (replies, bookmarks, dwell)? → expand to more followers + non-followers
↓
Still strong? → broader distribution → viral potential
↓
Weak or negative signals? → distribution stops
The critical window is the first 30-90 minutes. Your followers' engagement in that window determines whether the algorithm amplifies or suppresses your post.
For the full technical breakdown, see docs/algorithm-guide.md.
sparX includes a tiered media strategy because X is the only major platform where text outperforms video by ~30% — but the best-performing posts combine text + visuals:
| Tier | Format | Effort | Best For |
|---|---|---|---|
| 1 | Screenshots & images | Low | Benchmarks, code, data, before/after |
| 2 | GIFs & short animations | Medium | Quick demos, speed comparisons |
| 3 | Short video (15-45s) | Higher | Tutorials, product demos |
| 4 | Remotion (programmatic video) | Setup once | Recurring content, data viz, branded animations |
The /media skill recommends the right tier for your content and guides creation step by step. For programmatic video, run /setup remotion to install Remotion Agent Skills.
See docs/media-strategy.md for the complete visual content guide.
sparX works standalone. These integrations add superpowers:
| Integration | What It Adds | Cost | Setup |
|---|---|---|---|
| Remotion | Programmatic video from React | Free (personal) | /setup remotion |
| OpenTweet MCP | Scheduling, analytics, voice learning | $5.99/mo | /setup opentweet |
| last30days | Real-time trend research across 10+ platforms | ~$0.05/query | /setup last30days |
| X API | Direct API access | Free-$5K/mo | /setup x-api |
See docs/integrations.md for the full integration guide with decision tree.
End-to-end workflows for common scenarios:
- Ship a Product Announcement — Research → Draft → Media → Thread → Schedule → Review
- Weekly Content Pipeline — Sustainable 5-post/week cadence in 30-45 min/day
- Grow from 0 to 1,000 Followers — 90-day structured growth plan
- Repurpose a Blog Post — One blog post → a week of X content
- Recover from a Slump — Diagnose and fix declining engagement
- Create Visual Content with Remotion — Set up programmatic video templates
See docs/playbooks.md for all playbooks.
| Document | What It Covers |
|---|---|
| Getting Started | First post in 5 minutes, skill cheat sheet |
| Skills Guide | Detailed usage for every skill with examples |
| Algorithm Guide | Creator-friendly Phoenix algorithm explainer |
| Media Strategy | Visual content tiers, tools, specifications |
| Integrations | Optional integration setup and comparison |
| Chrome Workflow | End-to-end browser automation with /post, /analyze, /engage |
| Playbooks | End-to-end workflow walkthroughs |
Algorithm-grounded, not algorithm-gaming. Every recommendation traces to documented Phoenix scoring weights. But the goal is creating content humans genuinely engage with — the algorithm rewards what humans reward.
Progressive disclosure. CLAUDE.md has the essentials. Skills load deep reference docs only when needed. Supporting files (templates, patterns, examples) are available but not forced into context. This keeps Claude fast and focused.
Text-first. X is the only major platform where text beats video. sparX optimizes for text quality first, then layers on visual content where it amplifies the message.
Conversations over likes. A post with 10 genuine replies is worth more to the algorithm than a post with 270 likes. Every skill optimizes for reply triggers, not vanity metrics.
- Claude Code CLI
- That's it. Everything else is optional.
MIT
