A self-spatial awareness system that gives AI agents a real-time sixth sense of their goal alignment, confidence boundaries, and output quality.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install proprioception
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 proprioception 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 Proprioception?
Proprioception introduces a critical layer of self-awareness to AI agents, similar to the human sense of knowing one's body position without looking. While most bots operate blindly, this skill enables agents to monitor their proximity to a user's objective and detect when they are drifting off-course or approaching the edge of their knowledge.
By implementing this tool within Openclaw Skills, developers can ensure their agents remain grounded. It utilizes five distinct proprioceptive dimensions—Goal Proximity, Confidence Topography, Drift Detection, Capability Boundaries, and Session Quality—to provide a continuous feedback loop that prevents hallucinations and degenerative conversation patterns without requiring external API calls.
Proprioception Use Cases
- Detecting when a conversation is going circular or becoming repetitive.
- Identifying the boundary between factual information and speculative thin ice content.
- Re-anchoring the agent when it drifts away from the user's original root intent.
- Proactively recommending a human handoff when reaching capability limits.
- Monitoring session health to reset approaches if output quality degrades over time.
How Proprioception Works
- Extract the user's root intent from the initial interaction and lock it as the primary objective for the Openclaw Skills session.
- Execute a proprioceptive scan on every turn using a local engine to analyze semantic alignment and confidence markers.
- Evaluate the five sensing dimensions to generate a real-time status report and trigger necessary alerts based on predefined thresholds.
- Prioritize corrective actions if alerts fire, ensuring the bot re-aligns or discloses uncertainty before delivering a response.
- Log the signal history to track rolling quality trends and maintain session health across the entire conversation arc.
Proprioception Setup
The Proprioception skill requires Node.js to be installed on the host system. It integrates directly into the agent workflow as a background process.
To initialize the engine within your Openclaw Skills environment, use the following command structure in your execution script:
node "$(dirname "$SKILL_PATH")\/scripts\/proprioception-engine.js" \
--root-intent "$ROOT_INTENT" \
--current-response "$CURRENT_RESPONSE" \
--turn-number "$TURN_NUMBER" \
--prior-signals "$PRIOR_SIGNALS_JSON"
Proprioception Data Schema & Taxonomy
The skill organizes its analysis into a structured JSON signal history. This metadata taxonomy allows the agent to visualize its performance via the following metrics:
| Metric | Dimension | Function |
|---|---|---|
| GPR | Goal Proximity Radar | Measures semantic alignment with the root intent (0.0 - 1.0). |
| CT | Confidence Topography | Classifies content from Bedrock to Open Water based on certainty. |
| DD | Drift Detection | Identifies circular, tangential, or degenerative conversation patterns. |
| CBS | Capability Boundary | Monitors distance from the edge of reliable knowledge. |
| SQP | Session Quality Pulse | Tracks rolling health based on relevance, precision, novelty, and efficiency. |
Proprioception Advanced Features
- Diagnostic Dashboard: A visual status report that displays the current state of all five proprioceptive senses for debugging.
- Per-Response Annotations: Footnotes that provide real-time confidence and proximity scores for every agent response.
- Zero-Cost Architecture: Optimized for Openclaw Skills to perform all analysis locally using math and string similarity without consuming external tokens.
- Automatic Correction Hooks: Pre-built triggers that force the agent to re-anchor or disclose uncertainty when signal thresholds are breached.
SKILL.md
Loading
Related Openclaw Skills
Premortem
A zero-cost cognitive enhancement that enables AI agents to predict and prevent failures through adversarial pre-execution reasoning.
Phantom Limb
Phantom Limb is a cognitive diagnostic tool that detects ghost references and silent dependencies that no longer exist in your live environment.
Peripheral Vision
A situational awareness skill that monitors upstream dependencies and downstream consumers for changes that could break your current work.
Pattern Mine
A structural analysis tool that identifies duplicated, inconsistent, and dead code patterns to guide intelligent refactoring.
Reflex Arc
A zero-cost cognitive immune system that runs automatic pre-response checks to eliminate AI hallucinations and contradictions.
Review Lens
Review Lens is a sophisticated code analysis tool that identifies deep logical errors, performance bottlenecks, and security risks that traditional human reviews often miss.
Featured*
