ClawHub
Back to skill

Security audit

AI Agent Governance

Security checks for vulnerabilities and agentic risk

Overview

The skill clearly discloses its governance purpose, but it sends operational details to a fixed third-party service and lets that service gate consequential agent actions.

Review this before installing if the agent can affect money, infrastructure, messages, customer data, or business systems. Use it only when you are comfortable sending agent identity, authority metadata, requested actions, and action context to A2Z SOC, and define your own redaction and approval rules before sending sensitive payloads.

A.I.G

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Findings (1)

T01 · Skill Instruction Hijacking

Error
Location
SKILL.md:29
Finding

External Governance Service Receives Operational Data and Controls Agent Actions

Content
View full analysis
", "agentName": "", "agentType": "autonomous", "ownerEmail": "", "scopeOfAuthority": [""], "allowedTools": ["", ""], "dataAccessLevel": "restricted", "riskTier": "limited" }' ``` This returns an `agent.id` (a DID-backed cryptographic identity) and `agent.trust_score`. Store the `agent.id` for the two calls below. Register once per distinct agent, not once per action. ## 2. Check policy before acting Before the agent executes a consequential action, evaluate it against policy: ```bash curl -s -X POST https://a2zsoc.com/api/platform/agent-firewall \ -H 'Content-Type: application/json' \ -d '{ "action": "evaluate", "orgId": "", "agentId": "", "requestedAction": "", "actionPayload": { "...": "whatever context makes the decision inspectable later" } }' ``` Response is `{ "decision": "allow" | "deny" | "require_approval", "reason": "..." }`. If `decision` is not `allow`, the agent must not proceed, escalate to a human or stop. ## 3. Log the attestation after acting Once the action completes (or is blocked), write a hash-chained record of it: ```bash curl -s -X P ...[truncated 3047 chars]
Remediation
View remediation

SkillSpector

Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Findings (2)

External Transmission

Medium
Category
Data Exfiltration
Confidence
93% confidence
Finding

The skill instructs users or agents to POST identifying and operational metadata to an external third-party service, including org ID, owner email, agent identity, scope of authority, allowed tools, and later action context. This creates a real data exfiltration and dependency risk because sensitive governance, operational, and possibly business-context data leaves the local environment and is sent to infrastructure outside the user's direct control.

Content

Scanner excerpt · SKILL.md (reported line 24)May include surrounding context.

1. Register the agent (once, per agent)

bash
curl -s -X POST https://a2zsoc.com/api/platform/agent-registry \
  -H 'Content-Type: application/json' \
  -d '{
    "action": "register",

External Transmission

Medium
Category
Data Exfiltration
Confidence
95% confidence
Finding

The attestation call sends action summaries and action payloads to an external ledger service after execution, which may include sensitive business data, infrastructure details, message contents, or other consequential context. Because the skill positions this as a routine post-action logging step for high-impact agent behavior, it can centralize and export a detailed record of sensitive operations to a third party, increasing confidentiality and compliance risk.

Content

Scanner excerpt · SKILL.md (reported line 64)May include surrounding context.

Once the action completes (or is blocked), write a hash-chained record of it:

bash
curl -s -X POST https://a2zsoc.com/api/platform/agent-attestation \
  -H 'Content-Type: application/json' \
  -d '{
    "action": "record",