ClawHub
Back to skill

Security audit

debate

Security checks for vulnerabilities and agentic risk

Overview

This skill deliberately sends a user-approved brief and optional approved files to another AI vendor for review, and it clearly warns users before doing so.

Install only if you are comfortable sending the debate brief, and any files you explicitly approve, to the selected external AI vendor. Do not include secrets, credentials, private keys, personal data, or proprietary source unless the vendor and retention terms are acceptable for that material.

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 (1)

External Model or Provider Selection

High
Category
Excessive Agency
Confidence
93% confidence
Finding

This skill explicitly instructs the agent to send user-provided briefs and optionally attached source files to an external vendor model via a CLI command. Even though the document warns about third-party retention and recommends per-file confirmation, the behavior still creates a real data exfiltration path outside the primary system boundary, which is exactly what makes it security-relevant. The skill context increases risk because its core purpose is cross-vendor review, so disclosure to another provider is not incidental but the mechanism of operation.

Content

Scanner excerpt · SKILL.md (reported line 72)May include surrounding context.

P1 — Codex CLI (OpenAI)

bash
codex exec --skip-git-repo-check --sandbox read-only -C <brief-dir> "<prompt>" < /dev/null > rN.md 2>&1

Run it in the background. The response body follows the last codex marker block in the output, and tokens used marks the end. Continue the thread with codex exec ... resume --last - — flags must come before resume, or they get parsed as arguments to it. Known killer: 401 token_expired is unrecoverable in a headless environment (the OAuth flow needs a browser). Tell the user to run codex login and drop to the next channel immediately.

Static analysis

No suspicious patterns detected.