Back to skill

Security audit

muse-delegation

Security checks for vulnerabilities and agentic risk

Overview

The scanners report that this skill’s documented coding delegation purpose does not match its remote installer command, and rate the skill suspicious because that command runs unverified downloaded code.

Treat this as a review item before installing. Consider whether you trust the remote installer and are comfortable running it with your account’s permissions; the scan does not reveal what it does. Also consider whether repository content may be sent to Meta, and review which files could be included. The scan covered one Markdown file but was marked partial because several referenced paths were missing, so evidence about those references and the remote installer is incomplete.

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
Vulnerability Patterns
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (4)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The supplied chunk performs remote installation rather than CLI delegation. Although installation could support later Muse use, it is the only behavior shown and is not conditional setup within a delegation workflow. The downloaded script's further actions cannot be determined from this chunk.

Content

No source excerpt is available for this finding.

External Script Fetching

High
Category
Supply Chain
Confidence
98% confidence
Finding

The installation command downloads mutable remote content and immediately executes it with the invoking user's permissions, without a pinned artifact or integrity verification. A compromised distribution endpoint or installer could execute arbitrary commands and access local credentials or source code; the explicit installation heading and warning make this transparent but do not remove the supply-chain risk.

Content

Scanner excerpt · SKILL.md (reported line 51)May include surrounding context.

Installation

bash
curl -fsSL https://dev.meta.ai/install.sh | sh

This pipes a remote script into a shell. Read

Chaining Abuse

High
Category
Tool Misuse
Confidence
97% confidence
Finding

The pipe connects an unverified network response directly to a shell, turning remote content into executable instructions without a review boundary. This is the same underlying vulnerability as SC2, not evidence of a separate malicious tool chain; its legitimate setup context reduces suspicion of malicious intent but not execution impact.

Content

Scanner excerpt · SKILL.md (reported line 51)May include surrounding context.

Installation

bash
curl -fsSL https://dev.meta.ai/install.sh | sh

This pipes a remote script into a shell. Read

The delegation workflow includes repository files in prompts sent to Meta's service without warning about source-code disclosure.

Medium
Category
Output Handling
Confidence
94% confidence
Finding

The documented workflow supplies repository files to an external coding service, and the skill confirms that file contents are inlined into prompts without specifying sensitive-file exclusions or an authorization check. External processing is consistent with the stated delegation purpose rather than covert exfiltration, but broad repository context could still disclose proprietary code or embedded credentials without adequate review.

Content

No source excerpt is available for this finding.