CLI Reference

LSJI includes a comprehensive CLI for RL agents, LLM agents, runtime server, and plugins.

Installation

# Global install
npm install -g @game_ryo/lsji

# Or use npx
npx @game_ryo/lsji --help

---

RL Agent Commands (Legacy)

lsji train

Train the RL agent.

lsji train [options]

Option	Description	Default
--episodes <n>	Number of training episodes	200
--pattern <0-3>	Training pattern	0
--batch-size <n>	Database batch size	200
--opponent <type>	Opponent strategy	random
--storage <type>	Storage backend	sqlite
--db-path <path>	Database file path	./lsji.db
--alpha <n>	Learning rate	0.1
--gamma <n>	Discount factor	0.9
--epsilon <n>	Exploration rate	0.1
--json	Output as JSON	false

Training Patterns: 0=Random, 1=Always Rock, 2=Counter, 3=Sequential

lsji play

Play a single game against the RL agent.

lsji play --hand <0|1|2> [options]

Option	Description
--hand <0|1|2>	Your hand: 0=Rock, 1=Scissors, 2=Paper
--opponent <type>	Opponent strategy
--storage <type>	Storage backend
--db-path <path>	Database file path
--json	Output as JSON

lsji status

Show RL system status.

lsji status [options]

lsji start / lsji stop

Enable/disable RL training loop.

lsji start
lsji stop

---

LLM Agent Commands (v0.3+)

lsji agent run

Run an LLM agent on a task.

lsji agent run --task "Your task description" [options]

Option	Description	Default
--task <string>	Task description	(required)
--provider <name>	LLM provider: gemini, openai, anthropic, local	openai
--model <name>	Model name	gpt-4o-mini
--api-key <key>	API key (or use env var)	-
--max-cost <n>	Max cost per run (USD)	10
--max-tokens <n>	Max tokens per run	100000
--hitl <true|false>	Enable HITL approvals	true
--hitl-timeout <ms>	Approval timeout	300000
--conversation <tf>	Enable conversation memory	true
--semantic <tf>	Enable semantic memory	false
--episodic <tf>	Enable episodic memory	true
--hitl-required <list>	Comma-separated tools needing approval	file_write,api_call,send_email,code_exec
--max-steps <n>	Max agent steps	50
--json	Output as JSON	false

Examples:

# Simple task with Gemini
lsji agent run --task "Summarize TypeScript best practices" --provider gemini --model gemini-1.5-flash

# With budget limits
lsji agent run --task "Research AI trends" --max-cost 5 --max-tokens 50000

# Disable HITL for automation
lsji agent run --task "Read config file" --hitl false

lsji agent run-durable

Run with checkpointing (resumable after crash).

lsji agent run-durable --task "Your task" [options]

Option	Description	Default
--workflow-id <id>	Workflow identifier	auto-generated
--checkpoint-every <n>	Checkpoint every N steps	3
--resume-from <id>	Resume from checkpoint ID	-

Example:

# Durable run
lsji agent run-durable --task "Analyze codebase" --workflow-id my-analysis

# Resume after crash
lsji agent run-durable --task "Analyze codebase" --resume-from checkpoint_abc123

lsji agent status

Show agent configuration status.

lsji agent status [--json]

---

Budget Commands

lsji budget status

lsji budget status [--budgetId <id>] [--json]

lsji budget check

Check if operation is within budget.

lsji budget check --cost <n> --tokens <n> [--budgetId <id>] [--json]

lsji budget reset

Reset run budget.

lsji budget reset [--budgetId <id>]

---

HITL Commands

lsji hitl list

List pending approvals.

lsji hitl list [--limit <n>] [--json]

lsji hitl approve

Approve a pending request.

lsji hitl approve --id <approval-id> [--reason <string>] [--decider <name>] [--json]

lsji hitl reject

Reject a pending request.

lsji hitl reject --id <approval-id> --reason <string> [--decider <name>] [--json]

lsji hitl status

Show approval status.

lsji hitl status --id <approval-id> [--json]

---

Checkpoint Commands

lsji checkpoint list

List workflows with checkpoints.

lsji checkpoint list [--json]

lsji checkpoint show

Show checkpoints for a workflow.

lsji checkpoint show --workflowId <id> [--json]

lsji checkpoint recover

Recover from latest checkpoint.

lsji checkpoint recover --workflowId <id> [--json]

---

Idempotency Commands

lsji idempotency cleanup

Clean expired keys.

lsji idempotency cleanup

lsji idempotency list

List keys for an operation.

lsji idempotency list --operation <name> [--limit <n>] [--json]

---

Server Commands

lsji serve

Start the runtime server with web control panel.

lsji serve [options]

Option	Description	Default
--port <n>	Server port	3456
--no-ui	Disable web UI, API only	false
--storage <type>	Storage backend	sqlite
--db-path <path>	Database file path	./lsji.db

Examples:

lsji serve                    # Start with UI on :3456
lsji serve --port 8080        # Custom port
lsji serve --no-ui            # Headless API only

---

Plugin Commands

lsji plugin list

List loaded plugins.

lsji plugin list [--json]

lsji plugin create

Generate plugin template.

lsji plugin create --name <plugin-name>

lsji plugin load

Load plugins from directory.

lsji plugin load --path <directory>

---

Common Options

Option	Description	Default
--storage <type>	Storage: sqlite, better-sqlite, memory	sqlite
--db-path <path>	Database file path	./lsji.db
--json	Output as JSON	false
--help / -h	Show help	-

---

Environment Variables

Variable	Description	Default
GEMINI_API_KEY	Google Gemini API key	-
OPENAI_API_KEY	OpenAI API key	-
ANTHROPIC_API_KEY	Anthropic API key	-
OLLAMA_HOST	Ollama host URL	http://localhost:11434
LSJI_STORAGE	Default storage backend	sqlite
LSJI_DB_PATH	Default database path	./lsji.db
LSJI_SERVER_PORT	Server port	3456

---

Quick Reference Card

# RL (Legacy)
lsji train --episodes 500
lsji play --hand 0
lsji status --json

# LLM Agent
lsji agent run --task "Research topic" --provider gemini
lsji agent run-durable --task "Analyze" --workflow-id my-job

# Budget
lsji budget status
lsji budget check --cost 1 --tokens 5000

# HITL
lsji hitl list
lsji hitl approve --id abc --reason "OK"

# Server + UI
lsji serve
lsji serve --port 8080 --no-ui

# Plugins
lsji plugin list
lsji plugin create --name my-tools