ClawHub
Back to skill

Security audit

typed-decisions-around-llms

Security checks for vulnerabilities and agentic risk

Overview

This is a Markdown-only architecture guidance skill about safer typed decision layers around LLMs, with no hidden execution or data-access behavior found.

This skill is safe to install as guidance material. Users should still treat it as architectural advice, verify current product details in live documentation before implementation, and ensure any real automation built from the advice has explicit policy gates and human review for consequential actions.

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

SkillSpector

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

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · SKILL.md (reported line 100)May include surrounding context.

md
- Are they the same model call? Then the gate is decorative.

Watch for the disguised form: an LLM emits both a label saying an action needs
no approval *and* the cost estimate that the only remaining threshold compares
against. Both inputs to the gate come from the thing being gated.

## 4. The hybrid request path

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
85% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · SKILL.md (reported line 165)May include surrounding context.

md
review unavailable. Unfavorable or low-confidence ⇒ proposal-only, revision,
  or human. Only a complete favorable set makes the proposal *eligible* for the
  next policy gate.
- **A semantic review never auto-executes a consequential command.** It says
  what appears true of the supplied state; it does not grant permission.
- **Receipts go stale.** If the file, policy, or tool arguments changed, the
  earlier review is not current approval.

Static analysis

No suspicious patterns detected.