Back to skill

Security audit

Requirement Sharpening Stone

Security checks for vulnerabilities and agentic risk

Overview

This PRD-writing skill is mostly coherent, but it asks for under-scoped repository reading and default persistence/external document flows that users should review before installing.

Install only if you are comfortable with Feishu document creation, local PRD/log files, and possible web research. For confidential projects, choose local-only output, require explicit approval before any Feishu or WebSearch use, and do not allow repository inspection unless you provide narrow paths and review what will be copied into the PRD or logs.

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
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • 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
Findings (8)

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The skill claims it does not handle code or technical solutions, yet later phases require review against interface feasibility, field specs, error codes, API dependencies, and repo-derived functionality. This inconsistency can mislead users about the level of technical/system access and analysis the skill will perform, undermining informed consent and increasing the chance of oversharing sensitive engineering details.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The manifest examples are generic PM requests and insufficiently distinguish this skill from routine assistant behavior. Because the skill can create external documents and persist local artifacts, ambiguous triggers increase the risk of unintended side effects from ordinary conversation.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill states it is a pure process-orchestration PM skill, but its documented capabilities include local file read/write and later instructions to read existing code repositories to generate inheritance summaries. That creates a scope-expansion path where a non-technical requirements workflow can access source code and potentially expose sensitive implementation details unrelated to the user's immediate task.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill can create Feishu documents, perform web retrieval, and write local files, but the documentation does not clearly warn users that their content may be sent to external services or persisted locally. This creates a data-transmission and persistence risk, especially for confidential product plans, internal metrics, or proprietary materials.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The activation phrases are broad common-language requests such as '帮我写 PRD' and '跑完整流程', making accidental invocation plausible in normal conversation. Over-broad triggers are dangerous because they can initiate external actions, local writes, web retrieval, or data collection without the user realizing a specialized skill has been activated.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill defines per-stage logging to an experience-base directory but does not provide a clear upfront notice or consent flow for retention of project metadata and outcomes. Even if the schema looks operational, logging project names, stages, and outcomes can expose sensitive roadmap or business information and may violate user or organizational expectations.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The instruction to read existing code repositories for missing inheritance data is not clearly necessary for requirement sharpening and can cause unnecessary access to proprietary code. In a PM-oriented skill, this broadens data access beyond least privilege and may leak internal logic into documents, logs, or external outputs.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The learning-loop section prescribes comprehensive logging of raw user input, parsed intent, form-filled values, user modifications, compliance results, and AI-generated summaries. This is a broad data-collection pipeline that can capture sensitive business, personal, or regulated information and then resurface it in reports or knowledge bases, increasing exposure and secondary-use risk.

Content

No source excerpt is available for this finding.