Back to skill

Security audit

requirement-processor

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly a normal requirements workflow, but it asks the agent to persist project-derived notes inside the installed skill directory despite declaring only project/workspace document writes.

Install only if you are comfortable with the skill writing retrospective notes outside the current project. Before use, remove or override the instruction to update the skill's references directory, or require explicit approval and sanitization before any project information is persisted there.

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:137
Finding

Undeclared Persistent Write to the Installed Skill Directory

Content
View full analysis
此技能为纯文档/流程技能,不修改系统配置或执行外部命令。 ``` ### Technical Analysis The workflow instructs the Agent to update experience notes under the installed Skill's own `references/` directory. However, the permission declaration limits filesystem writes to the project working directory and states that writes are used to generate requirement documents. Writing to the Skill installation is outside that declared scope and violates least-privilege expectations. It also creates persistent state outside the audited project. Although the instruction does not explicitly request malicious content, credentials, or executable modifications, project-derived information written into a shared Skill directory could survive the current run and be consumed during later sessions. ### Attack Path 1. A user invokes the Skill for a project containing implementation details or other project-specific context. 2. The Agent reads the project as directed by the survey phase. 3. After implementation, the iteration phase directs the Agent to create or update experience notes under the Skill's `references/` directory. 4. The Agent writes project-derived content outside the declared working-directory write boundary. 5. The stored notes persist after the current task and may be read by subsequent invocations. 6. If insufficiently sanitized, the notes may expose project information across contexts or influence later Agent behavior. This path requires the Agent to possess write access to the installed Skill directory. The reviewed file does not itself establish such acce ...[truncated 730 chars]
Remediation
View remediation
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

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The manifest at L003 frames the skill as handling user requirements through analysis, documentation, phased planning, and iteration. But the workflow explicitly includes '开始编码实现' at L128, which goes beyond requirement processing into software implementation, a materially broader behavior than the stated description.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

L012 states '需求未澄清前不写代码' ('do not write code before requirements are clarified'), presenting a core intent of the skill. However, the documented workflow later includes '开始编码实现' at L128, meaning the skill is not purely a requirements/documentation processor and explicitly transitions into implementation work.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Low
Category
Not specified by scanner
Confidence
82% confidence
Finding

The stated purpose in L003 is requirement analysis, documentation, phased planning, and iterative closure for user needs. L137 directs updates to the skill's own references/ directory with experience notes, which is a self-maintenance capability not clearly implied by the manifest's user-facing requirement-processing scope.

Content

No source excerpt is available for this finding.