Tuitbot Skill

When to use

Use this skill when the user wants to:

• Grow their X (Twitter) account autonomously
• Discover relevant tweets and conversations in their niche
• Generate and post AI-powered replies, tweets, or threads
• Monitor target accounts and build relationships
• Review and approve queued posts (approval mode)
• View analytics on follower growth and content performance
• Configure posting limits, scoring thresholds, or brand voice
• Switch between Autopilot and Composer operating modes
• Manage drafts with AI-assisted content generation
• Browse the Discovery Feed and compose replies to scored tweets

Installation

cargo install --git https://github.com/your-org/tuitbot --bin tuitbot

Or build from source:

git clone https://github.com/your-org/tuitbot
cd tuitbot
cargo install --path crates/tuitbot-cli

Setup flow

1. Initialize configuration

tuitbot init --non-interactive

This creates ~/.tuitbot/config.toml with default values.

2. Configure settings

Set required values using tuitbot settings --set:

tuitbot settings --set business.product_name=YourProduct
tuitbot settings --set business.product_description="One-line description"
tuitbot settings --set business.product_keywords="keyword1, keyword2"
tuitbot settings --set business.industry_topics="topic1, topic2"
tuitbot settings --set llm.provider=openai
tuitbot settings --set llm.model=gpt-4o-mini

Secrets can also be set via environment variables:

export TUITBOT_LLM__API_KEY="sk-..."
export TUITBOT_X_API__CLIENT_ID="your-client-id"

3. Authenticate with X

tuitbot auth

This prints a URL. The user opens it in any browser (laptop, phone), authorizes, then pastes back the callback URL. Works on headless servers and VPS — no local browser needed. Tokens are saved to ~/.tuitbot/tokens.json.

Note: This step requires user interaction (visit URL, paste code). It cannot be fully automated.

4. Validate setup

tuitbot test --output json

Returns:

{"passed": true, "checks": [{"label": "Configuration", "passed": true, "message": "loaded from ~/.tuitbot/config.toml"}, ...]}

Available commands

All commands support --output json for machine-readable output.

tuitbot test --output json

Validate configuration, credentials, and connectivity.

{"passed": true, "checks": [{"label": "Configuration", "passed": true, "message": "..."}, {"label": "Business profile", "passed": true, "message": "..."}, {"label": "X API auth", "passed": true, "message": "..."}, {"label": "LLM provider", "passed": true, "message": "..."}, {"label": "Database", "passed": true, "message": "..."}]}

Exit code 1 if any check fails.

tuitbot settings --show --output json

Show current configuration as JSON (secrets redacted).

{"x_api": {"client_id": "abc", "client_secret": "***REDACTED***"}, "business": {...}, "llm": {"provider": "openai", "api_key": "***REDACTED***", ...}, ...}

tuitbot settings --set KEY=VALUE

Set a configuration value directly. Examples:

tuitbot settings --set scoring.threshold=70
tuitbot settings --set limits.max_replies_per_day=10
tuitbot settings --set approval_mode=true

tuitbot stats --output json

Show analytics dashboard.

{"follower_trend": [{"date": "2025-01-15", "follower_count": 1200, "following_count": 300, "tweet_count": 500}], "net_follower_change": 50, "top_topics": [{"topic": "rust", "format": "tip", "total_posts": 5, "avg_performance": 85.0}], "engagement": {"avg_reply_score": 73.5, "avg_tweet_score": 81.2}, "content_measured": {"replies": 25, "tweets": 18}}

tuitbot tick --require-approval

Force approval mode on for a single tick (queue posts for human review). This is automatically enabled when running inside OpenClaw (any OPENCLAW_* env var detected), but can also be set explicitly.

tuitbot run

Start the autonomous agent. This is a long-running process that runs 6 concurrent loops (discovery, mentions, content, threads, target monitoring, analytics).

# Start in background
nohup tuitbot run > /dev/null 2>&1 &

# Or with tmux
tmux new -d -s tuitbot 'tuitbot run'

Press Ctrl+C for graceful shutdown (30s timeout).

tuitbot discover --dry-run

Run one discovery cycle without posting. Useful for testing scoring.

tuitbot post --dry-run

Generate a tweet without posting.

tuitbot thread --dry-run

Generate a thread without posting.

MCP server

Tuitbot includes a built-in MCP (Model Context Protocol) server for structured tool access over stdio. This is the preferred integration method for AI agents — no CLI output parsing required.

Starting the server

tuitbot mcp serve

With a custom config path:

tuitbot -c /path/to/config.toml mcp serve

The server communicates via JSON-RPC 2.0 over stdin/stdout (newline-delimited).

Available MCP tools

Category	Tools
Analytics	get_stats, get_follower_trend
Action Log	get_action_log, get_action_counts
Rate Limits	get_rate_limits
Replies	get_recent_replies, get_reply_count_today
Targets	list_target_accounts
Discovery	list_unreplied_tweets
Scoring	score_tweet (6-signal engine)
Approval Queue	list_pending_approvals, get_pending_count, approve_item, reject_item, approve_all
Content Generation	generate_reply, generate_tweet, generate_thread (requires LLM provider)
Capabilities	get_capabilities (tier, rate limits, recommended max actions)
Config & Health	get_config, validate_config, health_check
Composer Mode	get_mode, compose_tweet, get_discovery_feed, suggest_topics

MCP vs CLI

• MCP: Typed JSON inputs/outputs, tool discovery via tools/list, no shell parsing. Best for programmatic agent integration.
• CLI with --output json: Simpler to invoke ad-hoc but requires spawning a process per command and parsing stdout.

When running inside OpenClaw via the plugin (plugins/openclaw-tuitbot/), MCP tools are automatically bridged into native OpenClaw tools prefixed with tuitbot_ (e.g., tuitbot_health_check, tuitbot_get_stats).

Approval workflow

When approval_mode = true, all generated content is queued for review instead of being posted immediately.

List pending items

tuitbot approve --list --output json

Returns:

[{"id": 1, "action_type": "reply", "target_tweet_id": "123", "target_author": "@user", "generated_content": "Great point!", "topic": "rust", "archetype": "AgreeAndExpand", "score": 85.0, "created_at": "2025-01-15T10:30:00Z"}]

Approve a specific item

tuitbot approve --approve 1 --output json

Returns:

{"id": 1, "status": "approved"}

Reject a specific item

tuitbot approve --reject 1 --output json

Returns:

{"id": 1, "status": "rejected"}

Approve all pending items

tuitbot approve --approve-all --output json

Returns:

[{"id": 1, "status": "approved"}, {"id": 2, "status": "approved"}]

Configuration reference

Key settings that can be modified via tuitbot settings --set:

Key	Type	Default	Description
business.product_name	string	""	Your product name
business.product_keywords	CSV	[]	Discovery keywords
business.industry_topics	CSV	[]	Content generation topics
llm.provider	string	""	AI provider: openai, anthropic, ollama
llm.model	string	""	Model name (e.g., gpt-4o-mini)
scoring.threshold	0-100	60	Minimum score to trigger a reply
limits.max_replies_per_day	int	5	Maximum replies per day
limits.max_tweets_per_day	int	6	Maximum tweets per day
limits.product_mention_ratio	0.0-1.0	0.2	Fraction of replies mentioning product
mode	string	autopilot	Operating mode: autopilot or composer
approval_mode	bool	false	Queue posts for human review
schedule.timezone	string	UTC	IANA timezone for active hours
schedule.active_hours_start	0-23	8	Start of posting window
schedule.active_hours_end	0-23	22	End of posting window

Environment variables use TUITBOT_ prefix with __ separator: TUITBOT_LLM__API_KEY, TUITBOT_SCORING__THRESHOLD.

Special env var: TUITBOT_APPROVAL_MODE=true|false overrides the approval_mode config value. When any OPENCLAW_* env var is present, approval mode is automatically enabled unless TUITBOT_APPROVAL_MODE=false explicitly opts out.

OpenClaw plugin

Tuitbot ships with an OpenClaw plugin at plugins/openclaw-tuitbot/ that bridges MCP tools into native OpenClaw tool registrations. When the plugin is loaded, approval mode is enabled by default for safety. See plugins/openclaw-tuitbot/openclaw.plugin.json for configuration schema.

Limitations

• tuitbot run is long-running: Start it as a background process. It does not return until stopped.
• tuitbot auth requires user interaction: OAuth 2.0 PKCE flow requires the user to visit a URL and paste back a code. Works on headless servers (VPS, SSH, OpenClaw) — no local browser needed — but cannot be fully automated.
• X API tier limits: X API access is pay-per-usage credits. Discovery, replies, mentions, target monitoring, tweets, and threads all require available credits (legacy Basic/Pro subscriptions may still exist for older accounts).
• Rate limits: The agent respects X API rate limits and configurable posting limits. It will not exceed them.