Back to skill

Security audit

Content Publish Lifecycle(内容发布全流程)

Security checks for vulnerabilities and agentic risk

Overview

Review is warranted because the release proofs overstate their protection and the file handling does not enforce its promised safety boundary.

Use a separate staging copy without symlinks and back up existing output files. Review the author and licensing statements before publication. Treat the chart as disclosed self-assessment, and do not rely on the bundled attestation as authenticated release evidence. The missing README leaves documentation incomplete; the inspected scripts show no exfiltration or automatic publishing.

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
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • 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 (3)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied code transforms precomputed security results into a chart. It neither runs the checks represented by those results nor implements the described design-to-publication stages. Although chart generation could support release documentation, no such integration is present in this chunk, and its standalone primary purpose materially differs from the declaration. File access is limited to local JSON input and SVG output; no network access or credential use is evident.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

File hashing and manifest verification could reasonably support the declared verification and release stages. However, the central output goes beyond supporting implementation details: it produces branded ownership assertions and restrictive legal statements without assessing authorship or rights. The supplied code also does not implement the advertised stage gates or lessons-learned capture. Its timestamps and hashes establish neither ownership nor independently trusted publication time. The mismatch concerns its materially different primary purpose and undisclosed legal-content generation, not network access or exfiltration.

Content

No source excerpt is available for this finding.

The generated attestation claims every package file is integrity-protected, but verification excludes the attestation itself.

Medium
Category
Excessive Agency
Confidence
98% confidence
Finding

The generated text claims that changing any package file changes its fingerprint, but SKIP_NAMES excludes ATTESTATION.md from both generation and verification. Someone able to modify the package can alter its human-readable ownership, licensing, or timestamp statements while verification still succeeds, undermining the release-evidence purpose of this tool.

Content

No source excerpt is available for this finding.