Back to skill

Security audit

主观题评分标准(初高中语英作文+语文阅读理解)

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed grading assistant that uses local rubric JSON files and does not show hidden code execution, credential access, exfiltration, or destructive behavior.

Install only if you want a Chinese curriculum-focused grading assistant. Review any proposed criteria/*.json updates before letting them persist, because saved rubric changes will affect later grading results.

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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (18)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The README states that AI批改时会“自动读取”并 in the workflow may also create corresponding scoring criteria, but it does not clearly warn users that teacher-provided input can result in persistent local file creation or modification. In a skill that accepts natural-language teacher instructions, this creates an unsafe trust boundary: unreviewed input can silently alter future grading behavior and poison the criteria store.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The evolution mechanism explicitly says teacher instructions will be automatically appended to criteria/*.json and take effect on later runs. That creates a persistent prompt/data poisoning vector: accidental, malicious, or malformed teacher input can silently corrupt grading standards, bias outputs, or break downstream logic that trusts these JSON files.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The description lists trigger words such as “打分” and “批改作业”, which are common phrases that could appear in everyday conversation and are not tightly scoped to this specific skill. The file also does not provide exclusion conditions or negative examples to clarify when the skill should not activate.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill states that newly added grading criteria in the criteria directory will be automatically read, but it does not prominently warn that this modifies persistent local files. In an agent environment, silent persistence can let ordinary conversation cause lasting state changes, creating integrity risks, configuration drift, and opportunities for prompt-injected or low-quality rubric data to be stored and reused later.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The workflow says the system may read or create corresponding grading criteria during scoring, but it does not make the persistent write behavior explicit to the user at the time of action. That means a user asking only for grading could unintentionally trigger durable configuration changes, which is dangerous because future grading behavior could be silently altered by prior inputs.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The evolution mechanism explicitly describes automatically appending teacher-supplied rules into criteria files for future automatic use, without a prominent safety gate. This creates a persistent-input channel where mistaken, malicious, or prompt-injected instructions can poison future grading standards, affecting integrity across subsequent sessions and users.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

This JSON manifest is entirely scoped to a Chinese junior-high exam context, including the name, grade, and exam type, but it does not explicitly document that the skill is region- or locale-specific by design. Under the policy, forcing a specific language/locale without user choice or clear justification is a natural-language policy concern.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The rubric states that using traditional Chinese characters (繁体字) may result in score deductions. This is a language/script policy constraint expressed in natural language, but the file does not offer any user choice or provide a clear region-specific justification for enforcing simplified-script norms.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The file declares a full score of 60, but the '二类上(59-62分)' range exceeds that maximum. In a grading skill, this inconsistency can cause invalid score outputs, unreliable ranking, or downstream logic errors if other components trust these ranges for normalization or report generation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The manifest names and describes the skill entirely in Chinese and is clearly tailored to Chinese-language junior secondary reading comprehension, but it does not explicitly state that this language/locale specialization is optional or region-specific. Under the language/locale policy rule, a fixed-language skill should either provide user choice or clearly document the justified regional scope.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

This JSON skill definition uses Chinese labels, descriptions, and grading guidance throughout, and does not indicate that users can choose another language or opt in to a Chinese-only experience. The policy requires flagging language or locale constraints when a specific language is effectively forced without user choice or explicit justification.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

This JSON file presents the skill name, description, scoring criteria, and instructional notes entirely in Chinese, including explicit pedagogical guidance, with no indication that users may choose another language or locale. Under the policy, forcing a specific language without user opt-in is a natural-language policy violation unless the locale constraint is clearly documented and justified.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

This JSON skill is explicitly defined as a Chinese-language high-school English writing rubric, with the name, grade, exam type, and description all fixed to a Chinese educational context. Under the stated policy, locale/language constraints should either offer user choice or be clearly documented as a justified region-specific tool; this file does not provide such opt-in or justification metadata.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
85% confidence
Finding

The skill metadata and interaction examples are entirely in Chinese, and the file does not indicate that users may choose another language for outputs or prompts. Under the language policy, a skill should not implicitly force a specific language unless it is documented as a justified locale-specific tool or offers opt-in choice.

Content

No source excerpt is available for this finding.

Vague Triggers

Low
Category
Not specified by scanner
Confidence
89% confidence
Finding

This JSON manifest describes the skill's content in very general terms but does not include any explicit invocation phrases, scope boundaries, or exclusion conditions. For manifest files, the absence of specific trigger constraints can make activation ambiguous and increase the chance of unintended invocation for general Chinese reading-comprehension requests.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
89% confidence
Finding

This manifest-like JSON file contains descriptions, prompts, scoring rules, and examples exclusively in Chinese, which can imply a forced language/locale constraint for users. The policy allows locale constraints when they are explicitly justified or offered as an opt-in, but no such explanation or choice appears in the file.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
96% confidence
Finding

L011-L035 explicitly defines four scoring dimensions, including splitting vocabulary/grammar into two separate dimensions. But the file description at L009 states there are three dimensions, and the official detailed section at L267 also describes three main aspects. This is an active documentation/content contradiction within the skill file about what rubric structure actually applies.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
98% confidence
Finding

The official scoring principles at L255 say the essay is scored in five levels (‘按五个档次给分’). However, the same section's five_levels array at L270-L347 actually contains six bands, including level 0 for ‘不得分’, and earlier L037-L140 also presents six levels. This is a direct inconsistency in the file's own stated scoring scheme.

Content

No source excerpt is available for this finding.