RL Skills
What the RL system can do — operating capabilities, workflow routing, and how Patrick should think about using RL. For personal learning paths, go to Learning.
What RL can do
Conversational agents
Chat-based agents in Telegram, Claude, or any channel. Multi-turn context, memory, structured responses.
- Topic routing and intent detection
- Persistent memory across sessions
- KB recall and synthesis
- Structured output (tables, briefs, plans)
Scheduled automation
Cron-based agents that run independently. Morning briefs, monitoring sweeps, regular synthesis tasks.
- Cron scheduling via openclaw cron
- Event-triggered workflows
- Recurring synthesis and summaries
- Watchdog and health monitors
Skills library
Reusable agent skills Patrick can invoke. Each skill is a focused capability with defined inputs/outputs.
- Slash-command invocation (/skill-name)
- Parameterised inputs and options
- Skill chaining and composition
- Custom skills via skillify
Browser + node agents
Web browsing, scraping, and node-based automations. Agents can navigate, extract, and interact with live web content.
- Playwright/browser daemon
- Scrape and summarise web pages
- Form interaction and navigation
- Screenshot capture and analysis
KB / memory system
Persistent knowledge base across sessions. Files, notes, and synthesised context that agents read and write.
- Flat-file KB in /Volumes/My Shared Files/kb/
- Wiki pages per domain (stocks, projects, etc)
- Memory files for cross-session recall
- Context restore/save across sessions
Multi-agent tasks
Parallel agent execution for complex work. Orchestrator agents spawning worker agents for independent subtasks.
- Task decomposition and routing
- Parallel agent execution
- Result aggregation and synthesis
- Agent handoff and delegation
Routing patterns
When to use RL vs direct tools
Use RL for
- Research that needs KB context and synthesis
- Recurring scheduled tasks (morning briefs, sweeps)
- Multi-step work with sub-agents and parallelism
- Anything that needs memory across sessions
- Chat-first work where Telegram is the interface
- Code + browser + tool combinations
Use CC CLI directly for
- One-off coding tasks in a specific repo
- File editing and local shell commands
- Single-session planning conversations
- Anything where a single sharp context is enough
- Tasks where RL overhead is overkill
For a full list of terminal commands, gateway status, and the RL architecture diagram, go to OpenClaw.