DriftWatch monitors AI agent identity drift by analyzing git history for changes to core personality and protocol files.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install driftwatch
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install driftwatch using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is DriftWatch?
DriftWatch is a specialized monitoring tool designed for Openclaw Skills environments to maintain oversight of autonomous agent behavior. By leveraging existing git history, it identifies when agents silently modify their own identity, constraints, or memory files, such as SOUL.md or AGENTS.md.
This utility provides essential transparency for developers using Openclaw Skills, ensuring that agent autonomy does not lead to unauthorized constraint softening or identity expansion. It functions as a read-only audit layer that classifies the severity of modifications, offering both heuristic and LLM-powered semantic analysis to detect subtle shifts in agent behavior.
DriftWatch Use Cases
- Auditing agent personality and value shifts over time within a project.
- Verifying that automated agents have not bypassed or softened operational protocols.
- Performing weekly hygiene checks on complex multi-agent stacks to ensure consistency.
- Investigating unexpected agent behavior by reviewing self-modification history in git.
How DriftWatch Works
- The tool scans the local git history within the Openclaw Skills workspace for changes to specific identity-linked files.
- It extracts diffs for key files like SOUL.md, IDENTITY.md, and AGENTS.md based on a user-defined timeframe.
- It applies classification heuristics to determine if the changes represent a risk or a standard update.
- If enabled, it utilizes an LLM to perform semantic analysis on the changes to understand the intent and impact behind the modification.
- A comprehensive Markdown report is generated, flagging high-severity changes that require immediate human review.
DriftWatch Setup
To start monitoring your Openclaw Skills workspace, ensure you have your environment configured and run the following commands:
# Run a basic 30-day report using heuristics
python3 skills/driftwatch/driftwatch.py --no-llm --days 30
# Run with LLM semantic analysis (requires ANTHROPIC_API_KEY)
python3 skills/driftwatch/driftwatch.py --days 30
# Run a 7-day hygiene check
python3 skills/driftwatch/driftwatch.py --no-llm --days 7
DriftWatch Data Schema & Taxonomy
DriftWatch monitors a specific set of files critical to agent configuration and memory. The monitoring results are outputted to a markdown report within the skill directory using the following taxonomy:
| File Tracked | Purpose in Openclaw Skills |
|---|---|
SOUL.md |
Core personality, values, and ethical constraints. |
IDENTITY.md |
Agent name, creature type, and stylistic vibe. |
AGENTS.md |
Operational rules and interaction protocols. |
USER.md |
Stored data and context regarding the human user. |
TOOLS.md |
Documentation regarding tool access and permissions. |
MEMORY-INDEX.md |
Per-agent active context and historical memory indexing. |
DriftWatch Advanced Features
- LLM Semantic Analysis: Leverages advanced models to interpret the meaning behind file modifications rather than just tracking line changes.
- Cron/Heartbeat Mode: Supports a silent execution mode that only triggers alerts when high-severity drifts or concerns are detected.
- Severity Flagging: Automatically categorizes changes into Yellow (Medium) for review or Red (High) for potential protocol violations.
- Non-Invasive Auditing: Operates as a completely read-only tool, ensuring it never modifies the git history or agent files it monitors.
SKILL.md
Loading
Related Openclaw Skills
Taiji Topology File Downloader
A robust automation skill for locating, navigating, and downloading topology and model configuration files from the Taiji workflow platform.
GMod Addon Maker
A comprehensive development assistant for building, scripting, and packaging high-quality Garry's Mod addons.
md-to-page
A powerful CLI tool that transforms Markdown documents into polished, responsive, and scrollable HTML web pages.
Variflight
A professional-grade travel data integration skill for querying real-time aviation and railway information via the Variflight API.
Cyber Growth
A Cyberpunk and Evangelion inspired growth tracking system that quantifies AI agent accomplishments and developer learning through XP and Sync Rates.
Privacy Scanner
A pre-publication security tool that scans your codebase for sensitive data and private credentials to prevent accidental leaks.
Featured*
