A self-evolving AI skill that turns any AI coding agent into a personalized X (Twitter) social media manager. Works with any niche, any goal, any follower count.
On first run, the agent interviews you about your identity, niche, goals, and voice — then researches your X account automatically and populates all strategy files. From that point on, it:
- Crafts posts with proven hook frameworks, character validation, and strategic research
- Writes replies in your authentic voice (not AI-sounding generic praise)
- Runs engagement sessions with Reply Cards for your approval before posting
- Plans content with a 60/40 engagement-to-showcase ratio
- Tracks performance and self-updates its own strategy files based on what's working
- Manages DM outreach for client acquisition, job hunting, or networking
- Operates the CLI — can post, reply, search, and analyze directly via
twitter-cli
- An AI agent that supports skills/instructions (see Installation below)
twitter-cli— installed automatically on first run viauv tool install twitter-cli- An active X (Twitter) session in your browser (for cookie-based auth)
The easiest way to install this skill is using Vercel's skills CLI, which automatically detects and configures all major coding agents (OpenCode, Claude Code, Cursor, Windsurf, Codex, Gemini CLI, etc.).
Run this single command in your terminal:
npx skills add rabden/X-twitter-social-manager-skillThe CLI will automatically prompt you to choose:
- Scope: Install globally (available in all workspaces at
~/) or project-scoped (restricted to the current workspace). - Agents: Choose which of your installed coding agents (OpenCode, Claude Code, Cursor, etc.) you want to enable the skill for.
- Method: Install via Symlink (recommended — allows instant git updates) or Copy.
If you prefer to install manually, copy the SKILL.md and references/ directory into your agent's config folder:
- OpenCode: Global (
~/.config/opencode/skills/x-social-manager/) or Project (.agents/skills/x-social-manager/) - Claude Code: Global (
~/.claude/skills/x-social-manager/) or Project (.claude/skills/x-social-manager/) - Cursor / Windsurf / Codex: Project (
.agents/skills/x-social-manager/) or Global
Place the SKILL.md and references/ directory in your agent's custom instructions path. The skill is self-contained — it only needs the agent to be able to read/write files, run bash commands, and search the web.
If your agent supports reading markdown files as instructions directly:
- Clone this repository into your workspace:
git clone https://github.com/rabden/X-twitter-social-manager-skill.git
- Tell your agent:
"Read SKILL.md and follow the instructions to set up X Social Media Manager." - The onboarding flow will handle the rest.
x-social-manager/
SKILL.md # Main orchestration — onboarding, workflows, rules
references/
user-profile.md # Your identity, niche, goals, voice (filled during onboarding)
memory.md # Living document — performance data, voice patterns, lessons
voice-anti-patterns.md # Banned AI phrases, verification checklist
real-replies-archive.md # Your actual replies word-for-word (auto-populated)
content-strategy.md # Content pillars, target accounts, DM templates
hook-library.md # 10 proven hook frameworks, ranked by performance
algorithm-playbook.md # X Phoenix algorithm mechanics, timing, growth phases
growth-strategy.md # 90-day tactical playbook (generated during onboarding)
branding-strategy.md # Positioning analysis, competitive gaps, lead funnel
content-planner.md # Post queue, scheduling logic
cli-reference.md # twitter-cli commands, flags, safety rules
subagent-definitions.md # System prompts for 4 specialized sub-agents
media/ # Downloaded images for analysis (gitignored)
When the skill detects onboarding_complete: false in user-profile.md, it runs a 3-phase setup:
- Interview — Asks about your handle, niche, expertise, goals, voice, constraints
- Research — Fetches your X profile, recent posts, real replies, and scans your niche landscape
- File Generation — Populates ALL reference files with your personalized data:
- Generates 4 content pillars for your niche
- Writes a 90-day growth playbook for your specific situation
- Extracts your voice patterns and reply archetypes from real data
- Identifies Tier 1/2/3 accounts to engage with
- Drafts starter posts and a content calendar
After onboarding, the skill works identically to a hand-customized version.
| # | Workflow | What It Does |
|---|---|---|
| 1 | Craft a Post | Research -> 2-3 variations with hooks -> post on approval |
| 2 | Craft a Reply | Voice-matched reply with image analysis -> approval -> post |
| 3 | Write a Thread | 3-7 tweet thread with per-tweet char counts |
| 4 | Engagement Session | Feed scan -> Reply Cards -> post approved replies |
| 5 | Content Planner | Feed scan + competitor analysis -> draft 3-5 posts -> schedule |
| 6 | Visual Post Analysis | Download + analyze images from any post |
| 7 | Strategy Chat | Data-grounded advice from your performance history |
| 8 | DM Outreach | Research target -> warm DM draft |
| 9 | Intelligence Update | Performance analysis -> auto-update strategy files |
The skill spawns 4 specialized sub-agents (if the host IDE supports them):
| Agent | Purpose |
|---|---|
x-reply-crafter |
Drafts voice-matched replies |
x-feed-researcher |
Runs research, produces Research Summaries |
x-post-composer |
Crafts post text from research |
x-intel-updater |
Analyzes performance, updates strategy files |
If sub-agents aren't supported, the main agent handles everything directly — nothing breaks.
- Voice authenticity — Replies are matched against your real writing patterns, not generic AI output. A banned-phrases list and verification checklist prevent AI-sounding text.
- Research-first — Every post requires a mandatory Research Summary before drafting. No shooting from the hip.
- Approval-gated — The agent never posts without explicit user approval.
- Self-evolving — The Intelligence Update workflow re-ranks hooks, updates strategy, and profiles your audience based on real performance data.
- Algorithm-aware — All strategies are built on the X Phoenix algorithm's actual ranking signals (Author-Engaged Reply > Bookmark > Like).
MIT