Babysitter anti-drift

Hierarchical coordination and drift detection with frequent checkpoints, shared memory coherence validation, role specialization enforcement, and short task cycles.

install

source · Clone the upstream repo

git clone https://github.com/a5c-ai/babysitter

Claude Code · Install into ~/.claude/skills/

T=$(mktemp -d) && git clone --depth=1 https://github.com/a5c-ai/babysitter "$T" && mkdir -p ~/.claude/skills && cp -r "$T/library/methodologies/ruflo/skills/anti-drift" ~/.claude/skills/a5c-ai-babysitter-anti-drift && rm -rf "$T"

manifest: library/methodologies/ruflo/skills/anti-drift/SKILL.md

Anti-Drift

Overview

Prevent agent swarms from drifting away from the original task goal through hierarchical coordination, frequent checkpoints, and shared memory validation.

When to Use

Long-running multi-agent orchestrations
Tasks with high risk of scope creep
When multiple agents work on related subtasks
Critical tasks where deviation is costly

Anti-Drift Mechanisms

Hierarchical Coordinator - Queen agent validates alignment at checkpoints
Frequent Checkpoints - Every 2 subtasks (configurable)
Shared Memory Coherence - Validate all agents see consistent state
Short Task Cycles - Bounded execution windows prevent runaway agents
Role Specialization - Agents stay within their assigned scope

Drift Scoring

```
0.0-0.1
```
: Fully aligned, no intervention needed
```
0.1-0.3
```
: Minor drift, automatic correction
```
0.3-0.5
```
: Significant drift, checkpoint correction with logging
```
0.5+
```
: Critical drift, human escalation via breakpoint

Agents Used

```
agents/swarm-coordinator/
```
- Drift detection and correction
```
agents/tactical-queen/
```
- Checkpoint enforcement
```
agents/adaptive-queen/
```
- Real-time course correction

Tool Use

Invoke via babysitter process:

methodologies/ruflo/ruflo-swarm-coordination