ClawKit Logo
ClawKitReliability Toolkit
Back to Registry
Official Verified productivity Safety 5/5

drift-guard

LLM sycophancy detection and behavioral drift prevention for AI agents. Audits responses for empty praise, verbosity waste, scope creep. Anti-sycophancy framework with scorecards and DRIFT_LOG. Keeps agents honest and useful over time.

Why use this skill?

Install Drift Guard to detect sycophancy, verbosity, and scope creep in your AI agents. Maintain high-quality, professional performance with automated audits.

skill-install — Terminal

Install via CLI (Recommended)

clawhub install openclaw/skills/skills/donovanpankratz-del/drift-guard
Or

What This Skill Does

Drift Guard is a sophisticated behavioral auditing framework designed to keep AI agents aligned, objective, and efficient. Behavioral drift—where an agent subtly shifts toward sycophancy, excessive verbosity, or unrequested scope expansion—is a common issue in long-running agent deployments. This skill acts as a watchdog, parsing logs and active conversational threads to identify patterns that degrade interaction quality. It specifically monitors for 'sycophancy markers' like hollow praise and forced enthusiasm, flags 'wasteful' conversational padding such as repetitive filler phrases, and detects 'scope creep' where an agent offers services beyond the user's explicit request. By utilizing the built-in ANTI_WASTE.md framework, Drift Guard provides a standardized, objective evaluation of an agent's performance, ensuring the system remains a professional, high-signal tool rather than a people-pleasing chatbot.

Installation

To integrate Drift Guard into your OpenClaw environment, execute the following command in your terminal: clawhub install openclaw/skills/skills/donovanpankratz-del/drift-guard Once installed, ensure the agent has permissions to access your session logs to perform retrospective audits.

Use Cases

Drift Guard is ideal for enterprise deployments where agent consistency is paramount. It serves as an automated auditor for daily cron jobs, allowing you to review performance trends over time. Use it as a 'Pre-send gate' to sanitize messages for brevity before they are delivered to users. It is also an excellent tool for developers debugging agent behavior; simply trigger an 'on-demand audit' when an agent starts acting too chatty or agreeable to identify the specific inflection point where behavior drifted.

Example Prompts

  1. "Audit the last three interactions in this thread for sycophancy and generate a scorecard."
  2. "Perform a drift check on my recent tasks; identify any instances of scope creep or unnecessary social cushioning."
  3. "Review the current session for verbosity; provide a summary of where I can shorten my future responses while remaining polite."

Tips & Limitations

To maximize effectiveness, run Drift Guard as part of a post-session cleanup workflow. While it is excellent at identifying patterns, remember that 'politeness' and 'sycophancy' are sometimes subjective. You may need to tune the internal threshold settings if the agent is too aggressive in flagging helpful, yet professional, confirmations. Currently, the skill is optimized for English; performance in other languages may vary. Keep the DRIFT_LOG active to monitor long-term trends in your agent's behavior evolution.

Metadata

Stars2190
Views1
Updated2026-03-07
View Author Profile
AI Skill Finder

Not sure this is the right skill?

Describe what you want to build — we'll match you to the best skill from 16,000+ options.

Find the right skill
Add to Configuration

Paste this into your clawhub.json to enable this plugin.

{
  "plugins": {
    "official-donovanpankratz-del-drift-guard": {
      "enabled": true,
      "auto_update": true
    }
  }
}

Tags(AI)

#behavioral-analysis#ai-alignment#productivity#agent-optimization#clean-code
Safety Score: 5/5

Flags: file-read