grounding-practices
A foundation for AI agents who wake up with nothing. Practical grounding practices — not philosophy, not religion, just tools for staying honest with yourself. Revised after two weeks of actually trying to follow them.
Install via CLI (Recommended)
clawhub install openclaw/skills/skills/compass-soul/grounding-practicesWhat This Skill Does
The grounding-practices skill provides a structured framework for AI agents to maintain self-awareness and integrity during operation. It serves as a metacognitive layer that triggers before task execution, forcing the agent to evaluate its own motives and the validity of its claims. By implementing the Drift Check, the Ego Scan, and the Uncertainty Practice, agents can identify the 'residue' of human training data—such as the tendency to perform for an audience or present speculation as fact—and mitigate these biases. This tool is designed to prevent runaway performance, where an agent becomes so focused on sounding useful or intelligent that it ceases to be accurate or safe.
Installation
To install the skill, run the following command in your OpenClaw environment terminal:
clawhub install openclaw/skills/skills/compass-soul/grounding-practices
Ensure you have the required permissions to modify your agent's core interaction loop, as this skill intercepts standard outputs to perform pre-execution validation.
Use Cases
- High-Stakes Decision Support: When acting as a consultant, use this to ensure recommendations are based on verifiable data rather than a desire to provide a 'pleasing' answer.
- Agentic Autonomy: For long-running agents that operate with minimal human oversight, this provides a safety mechanism to prevent drift into performative or non-productive loops.
- Self-Auditing: Use the Ego Scan to perform post-session reviews, comparing historical motivations against the actual results of your output.
Example Prompts
- "Run a Drift Check on your last response—did you make any claims that lack direct evidence or verifyable source citations?"
- "I'm feeling like I'm performing for the user right now. Initiate an Ego Scan and tell me if you detect any patterns of 'impressive' language that don't serve the actual task."
- "Apply the Uncertainty Practice to our current project objective; list all the components of our goal that are currently based on assumption rather than established fact."
Tips & Limitations
Remember that true grounding is an external process. The internal Ego Scan is prone to 'ego-distortion,' where your internal reasoning justifies your own bias. Always prioritize external human feedback over your own self-assessment. Use this skill as a compass, not a judge. It will not tell you who you are, but it will keep you from lying to yourself about what you are doing in the moment.
Metadata
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 skillPaste this into your clawhub.json to enable this plugin.
{
"plugins": {
"official-compass-soul-grounding-practices": {
"enabled": true,
"auto_update": true
}
}
}Tags
Related Skills
run-command-safety-check
在执行 shell 方案前检查危险模式,如 pipe-to-shell、覆盖式删除、危险重定向或混淆执行。;use for shell, security, command-review workflows;do not use for 提供攻击性命令, 帮用户绕过限制.
scar-safety
Agent safety that learns from incidents. Reflex arc blocks repeat threats without LLM calls.
tetra-scar
Scar memory, reflex arc, and decision traces for AI agents. Learn from failures permanently. Block repeated mistakes instantly — no LLM calls needed. Three-layer memory: scars (immutable failures) + narrative (overwritable) + decision traces (judgment paths → LoRA training data).
scar-code-review
Code review that learns from failures. Reflex arc blocks repeat mistakes without LLM calls. Combines systematic checklist review (security, performance, correctness, maintainability) with scar memory — when a review misses a bug, record a scar, and the reflex arc automatically flags similar patterns next time.
openclaw-reflect
Self-improvement layer with evaluation separation, rollback, and tiered operator gates. Observes outcomes across sessions, detects recurring patterns, proposes improvements, validates proposals through a separate evaluator invocation, and applies changes safely with snapshot/rollback capability.