Back to skill

Security audit

Meihua Yishu — 梅花易数占卜

Security checks for vulnerabilities and agentic risk

Overview

This is a mostly coherent Chinese divination skill, but it automatically reads another skill's persistent memory without clear field limits or fresh user consent.

Install only if you are comfortable with a divination skill that may reuse stored fortune-telling preferences and may read a fortune-hub memory file. Prefer a version that asks before reading cross-skill memory and limits reads to specific preference fields.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • 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)

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
SKILL.md:64
Finding

Unnecessary Cross-Skill Access to Persistent User Memory

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 64–83
Vulnerability Type: Cross-Skill persistent-memory access beyond the minimum required scope
Risk Level: Medium

Evidence

The following is an English translation of the relevant Skill instructions:

markdown
1. `MEMORY.md` in this directory — use first (stores the user's divination preferences)
2. `fortune-hub/MEMORY.md` in the same repository, if present — supplement shared basic fields

If usable data exists, use it directly without asking again.

**Write**: Write to this directory's `MEMORY.md` only when the user explicitly states a preference:

# User Preferences

## Divination Preferences
- Default divination method: time / number / text
- Interpretation style: concise / detailed

**Update**: Update the corresponding field when the user requests a preference change.

Technical Analysis

The Skill instructs the agent to inspect both its own persistent memory and the memory owned by another Skill. Its deterministic calculation only requires user-provided numbers, text, or the current time, so reading fortune-hub/MEMORY.md is not necessary for the core task.

The phrase “shared basic fields” does not define a field allowlist, sensitivity restrictions, a consent boundary, or a purpose limitation. The instruction to use available data “without asking again” may consequently cause information collected in another context to be accessed and incorporated without the user's informed consent.

The local write behavior is narrower because it is limited to explicitly expressed preferences. The reviewed instructions do not direct the agent to write attacker-controlled rules, executable content, or hidden instructions into memory. Therefore, the issue is best classified as a least-privilege violation rather than agent memory poisoning.

Attack Path

  1. A user invokes the Meihua Yishu Skill.
  2. The agent follows SKILL.md and che ...[truncated 1334 chars]
Remediation
View remediation

Remediation Suggestions

  1. Remove the automatic read of fortune-hub/MEMORY.md unless cross-Skill state is essential to the requested operation.
  2. If cross-Skill access must remain, define a strict allowlist limited to:
    • Default divination method.
    • Preferred interpretation detail level.
  3. Require explicit user consent before every cross-Skill memory read.
  4. Inform the user which file and fields will be accessed before reading them.
  5. Reject unrelated, sensitive, instructional, or executable content found in shared memory.
  6. Treat memory values strictly as untrusted data and never as agent instructions.
  7. Preserve the existing explicit-consent requirement for writes, and clearly disclose the retention scope and deletion procedure.
  8. Apply separate namespaces and access controls so each Skill can access only its own persistent state by default.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (4)

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
82% confidence
Finding

整个技能说明、触发词、输出示例与交互文案均默认采用中文,没有说明是否支持其他语言,也未提供用户选择语言的机制。按照语言/locale 政策,这可能构成未获用户选择即强制特定语言的自然语言策略问题。

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The trigger list includes broad, everyday phrases such as “帮我起一卦” and especially “算一卦,” which can plausibly appear in ordinary conversation and cause unintended activation. In an agent setting, overbroad activation can route unrelated user input into this skill, leading to confusing behavior, incorrect responses, or bypass of more appropriate domain-specific skills.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The guide is entirely written to drive Chinese-language interpretation output and does not indicate any user language preference handling or opt-in. This can cause the skill to ignore the user’s requested language, reducing usability, accessibility, and informed consent, though it is not a direct code-execution or data-exfiltration risk.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

This markdown file contains user-facing instructional content solely in Chinese from the title onward. Under the language/locale policy, forcing a specific language without user opt-in can be a policy concern unless the locale restriction is explicitly documented and justified.

Content

No source excerpt is available for this finding.