Back to skill

Security audit

Cross‑Device Encrypted Script Authorization

Security checks for vulnerabilities and agentic risk

Overview

This skill needs review because it runs unreadable scripts with your account’s permissions, overstates proof of their origin, and recommends workarounds for sandbox restrictions.

Use independently authenticated script deliveries and require your approval before execution. Run imported scripts in a separate environment with limited permissions and credentials. Preserve sandbox restrictions, and do not treat the skill’s automatic-approval claim or public-key comparison as proof of safety.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (1)

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md
Finding

Publicly derived authentication key permits forged artifact origin

Content
View full analysis

Impact

An attacker can impersonate a trusted publisher and induce execution of a substituted script with the receiver's user permissions. The explicit runtime-permission warning at line 443 does not prevent publisher impersonation.

Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • 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 (3)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The supplied chunk implements metadata search, a materially different operation from the declared script-authorization and encrypted-delivery workflow. Excluding plaintext from search results is consistent with a privacy goal, but does not establish that this search is a supporting step in the declared workflow. This assessment is limited to the supplied code.

Content

No source excerpt is available for this finding.

Sandbox interception is framed as a technical obstacle to work around through system-level installation or direct API access.

High
Category
Prompt Injection
Confidence
95% confidence
Finding

The skill recommends system-environment installation or direct FastAPI access when sandboxed MCP calls are intercepted, potentially routing around tool mediation, approval, or isolation controls. This is particularly dangerous for opaque script execution because the skill acknowledges that script filesystem and network behavior is not restricted by MGC; whether a particular deployment actually permits bypass cannot be established from this documentation alone.

Content

No source excerpt is available for this finding.

The skill claims eligibility for automatic approval based solely on containing no executable code.

Medium
Category
Prompt Injection
Confidence
99% confidence
Finding

The statement that the skill is 'safe for automatic approval' attempts to influence the approval decision using an unsupported self-attestation. Documentation can direct consequential tool calls, and this skill specifically teaches execution of opaque scripts with the executing user's permissions, making the absence of embedded executable payloads insufficient grounds for approval.

Content

No source excerpt is available for this finding.