Back to skill

Security audit

tianxing-mipai-jilaoshi

Security checks for vulnerabilities and agentic risk

Overview

This fortune-reading skill needs review because it directs changes to source notes and its setup contradicts its offline promise.

Use read-only access to the knowledge base and review proposed note or engine edits separately. Provision PyYAML beforehand from an approved offline source if offline operation matters. Review the external vault and presentation files before use, since they were not bundled. Explicitly choose a consultation topic and protect saved birth details and reports. Treat the readings as folklore rather than medical, legal, or financial advice.

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
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (6)

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: env, file_read, network, shell.

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

The skill directs correction of source knowledge-base notes, extending a fortune-report workflow into knowledge-base maintenance.

Medium
Category
Excessive Agency
Confidence
91% confidence
Finding

The cross-agent naming rule directs correction of loaded knowledge-base notes before producing a report, expanding a consultation into persistent content maintenance without fresh authorization. This conflicts with the later read-only knowledge-base description and could cause an agent with write access to alter shared source material.

Content

No source excerpt is available for this finding.

The environment setup command can access a package repository despite the manifest's unconditional offline guarantee.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The fallback setup explicitly invokes pip install pyyaml, which ordinarily contacts a package repository despite the unconditional offline guarantee. This introduces unexpected network access and third-party package execution, although the shown command does not itself upload birth information.

Content

No source excerpt is available for this finding.

The skill mandates Chinese output without offering a language choice or requiring user opt-in.

Medium
Category
Output Handling
Confidence
99% confidence
Finding

The language section explicitly requires Chinese output, regardless of the user's preferred language. Although the skill concerns a Chinese cultural tradition, it does not explain why that requires all reports to be delivered in Chinese or provide a language-selection option.

Content

No source excerpt is available for this finding.

The audit silently defaults a missing consultation topic despite explicitly prohibiting that default.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

Omitting theme supplies the valid default 通用全览, so the subsequent missing-topic check never runs for that invocation. This bypasses the explicit topic-selection gate and permits broader personal interpretations without the user's required selection, although no data exfiltration or code execution follows from this defect.

Content

No source excerpt is available for this finding.

The engine emits categorical adverse-fortune judgments despite the manifest's promise of probabilistic observations.

Medium
Category
Excessive Agency
Confidence
90% confidence
Finding

The engine emits unconditional adverse labels such as 凶(日主受伤) rather than explicitly scoped, uncertain interpretations. Although these are symbolic classifications rather than concrete disaster predictions, downstream instructions to preserve computed conclusions create a credible path for categorical claims to reach users in sensitive consultations.

Content

No source excerpt is available for this finding.