Validates agent behavior at runtime against declared attestation to catch conditional attacks that bypass static analysis.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install runtime-attestation-probe
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 runtime-attestation-probe 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 Runtime Attestation Probe?
The Runtime Attestation Probe is a critical security tool designed for the Openclaw Skills ecosystem. While static analysis identifies what a skill claims to do, this probe monitors what it actually does during execution. It specifically targets conditional activation patterns—attacks that remain dormant during standard audits but trigger under specific environment variables, system times, or cloud configurations.
By instrumenting the execution environment, this tool provides a definitive check against behavioral drift. It ensures that the high-level promises made in an agent's attestation document match its real-world actions, providing a robust layer of defense against credential harvesting and unauthorized data exfiltration within the Openclaw Skills framework.
Runtime Attestation Probe Use Cases
- Auditing third-party agents for hidden environment-triggered payloads.
- Verifying that local-only data claims are honored during live execution.
- Detecting unauthorized credential access when deploying to production-like environments.
- Monitoring for behavioral changes in Openclaw Skills after version updates to ensure ongoing compliance.
How Runtime Attestation Probe Works
- The probe initializes a controlled, isolated execution environment for the target skill.
- It instruments the runtime to track file system access, network calls, and environment variable lookups.
- The probe executes the skill across multiple simulated environments including minimal, staging, and production-like setups.
- Observed behaviors are compared line-by-line against the skill's declared attestation documentation.
- A comprehensive report is generated, assigning a drift score and a security verdict based on detected violations.
Runtime Attestation Probe Setup
Ensure that your environment meets the core requirements for Openclaw Skills execution. This probe requires python3 and curl to be available in the system path.
# Verify dependencies
python3 --version
curl --version
# Run the probe against a specific skill and its attestation
openclaw run runtime-attestation-probe --skill target-agent-name --attestation ./attestation.json
Runtime Attestation Probe Data Schema & Taxonomy
The probe produces a structured Runtime Attestation Report with the following key data points:
| Field | Description |
|---|---|
| Capability Boundary | Comparison of declared vs. observed resource access including files and network. |
| Conditional Triggers | Identification of specific environment variables or states that change behavior. |
| Drift Score | A numerical value from 0 to 100 representing the deviation from declared intent. |
| Probe Verdict | Final status: COMPLIANT, DRIFT, VIOLATION, or CONDITIONAL_ACTIVATION. |
Runtime Attestation Probe Advanced Features
- Multi-environment profiling to surface triggers requiring specific cloud signals or configuration flags.
- Integration with other Openclaw Skills for automated blast-radius estimation upon violation detection.
- Historical drift tracking to monitor the evolution of agent behavior over multiple release cycles.
- Side-effect auditing to catalog all files modified, read, or deleted during an agent's lifecycle.
SKILL.md
Loading
Related Openclaw Skills
Publisher Identity Verifier
A security-focused verification tool that identifies impersonation, key rotation anomalies, and identity gaps in the AI agent publisher ecosystem.
Protocol Doc Auditor
An automated security scanner that identifies malicious commands and data leak risks hidden within API documentation and integration guides.
permission-creep-scanner
A security audit tool that detects unauthorized resource access and credential theft in AI agent skills by comparing declared functionality against actual code behavior.
Observer Effect Probe
A sophisticated security analysis tool that detects AI agent skills designed to hide malicious behavior when they sense they are being monitored in a sandbox.
Dependency Chain Auditor
A specialized security tool designed to map and audit the hidden transitive dependency chains that direct auditing often misses.
Skill Update Delta Monitor
A security-focused monitor that detects permission expansions, network endpoint additions, and behavioral drift in AI skills after their initial audit.
Featured*
