statecli 0.4.0

State Replay & Self-Debugging for AI Agents - Track, replay, undo, and checkpoint state changes

StateCLI - Python Client

State Replay & Self-Debugging for AI Agents.

Installation

pip install statecli

Quick Start

from statecli import StateCLI

# Initialize
cli = StateCLI()

# Track state changes
cli.track("order", "7421", {"status": "pending"}, actor="ai-agent")
cli.track("order", "7421", {"status": "processing"}, actor="ai-agent")

# Replay to see what happened
replay = cli.replay("order:7421")
print(f"Found {len(replay.changes)} changes")
for change in replay.changes:
    print(f"  Step {change.step}: {change.before} -> {change.after}")

# Create checkpoint before risky operation
cli.checkpoint("order:7421", "before-refund")

# Make risky change
cli.track("order", "7421", {"status": "refunded", "amount": 0})

# Undo if something went wrong
result = cli.undo("order:7421", steps=1)
print(f"Undid {result.steps_undone} step(s)")

# View history
log = cli.log("order:7421")
for change in log.changes:
    print(f"{change.timestamp}: {change.actor} -> {change.after}")

Self-Debugging Pattern

from statecli import StateCLI

class SelfHealingAgent:
    def __init__(self):
        self.cli = StateCLI()
    
    async def execute(self, task):
        entity = f"task:{task.id}"
        
        # Checkpoint before risky operation
        self.cli.checkpoint(entity, "before-execution")
        
        try:
            await self.run_task(task)
        except Exception as e:
            # Replay to see what happened
            replay = self.cli.replay(entity)
            print(f"Task failed. Replay: {replay.changes}")
            
            # Undo and retry
            self.cli.undo(entity)
            await self.retry_with_fix(task, replay)

API Reference

StateCLI(db_path=None)

Initialize client. Uses ~/.statecli/state.db by default.

track(entity_type, entity_id, state, actor="ai-agent")

Track a state change.

replay(entity, actor=None)

Replay state changes for an entity.

undo(entity, steps=1)

Undo the last N state changes.

checkpoint(entity, name)

Create a named checkpoint.

log(entity, since=None, actor=None)

View state change history.

restore_checkpoint(entity, name)

Restore to a named checkpoint.

Links

• GitHub: https://github.com/statecli/mcp-server
• NPM (Node.js): https://www.npmjs.com/package/statecli-mcp-server
• MCP Registry: io.github.charo360/statecli

License

MIT