Back to skill

Security audit

GStack Agent

Security checks for vulnerabilities and agentic risk

Overview

This is a software-delivery helper, but it can automatically change code, push branches, and deploy production from broad natural-language requests without enough confirmation boundaries.

Install only if you are comfortable with an agent that may make code changes and perform release operations. Use explicit slash commands, prefer /qa-only for read-only testing, and require manual confirmation before any git push, PR merge, deployment, revert, or update to agent instruction files.

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
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (9)

Missing User Warnings

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The /land-and-deploy workflow performs merge-to-main, deployment, production smoke testing, and potential revert, yet it is triggered by broad phrases like '上线'. This is especially dangerous because the skill context is an end-to-end software factory with operational roles, making accidental production-impacting actions more plausible and severe.

Content

No source excerpt is available for this finding.

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
70% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · SKILL.md (reported line 356)May include surrounding context.

text
rm -rf, DROP TABLE, TRUNCATE, DELETE FROM(无 WHERE),
git push --force, git reset --hard, 
kubectl delete, terraform destroy

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
65% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · SKILL.md (reported line 356)May include surrounding context.

text
rm -rf, DROP TABLE, TRUNCATE, DELETE FROM(无 WHERE),
git push --force, git reset --hard, 
kubectl delete, terraform destroy

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

文件整体以中文描述触发短语和使用方式,相当于将技能交互入口固定在单一语言环境中,但未说明是否支持用户选择语言,也未给出必须限定中文的区域性理由。按组织语言/地区策略,这属于缺少用户语言选择的自然语言约束。

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The top-level skill description says the skill triggers on broad everyday phrases like '做代码审查', '跑 QA', '发布代码', and '写文档'. Because these phrases overlap with ordinary user requests, the skill may activate high-capability workflows unintentionally, including code modification or release-related behavior, without an explicit mode-selection step.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

Many role sections use broad natural-language triggers such as '帮我测试', '代码有问题吗', '准备发布', or '上线', without strong boundaries. In this skill, those triggers map to roles that may run tests, alter code, push branches, or deploy, so ambiguous activation increases the chance of unintended privileged actions.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The /review workflow states it will '自动修复' P0 and some P1 issues, meaning code can be changed as part of a review flow. Users asking for analysis may reasonably expect comments only, so silent transition from review to modification creates integrity and trust risks, especially if the wrong files or logic are changed.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The /qa mode says discovered defects will be automatically fixed and accompanied by regression tests. QA is commonly expected to validate behavior, not mutate the codebase, so this creates a substantial risk of unintended changes during a testing request.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The /ship workflow includes rebasing, running tests, adding tests, pushing a branch, and creating a PR, but the section lacks a prominent warning that these are repository-mutating remote actions. In practice, a user could invoke a seemingly routine 'ship' action and unintentionally publish changes or rewrite local branch state.

Content

No source excerpt is available for this finding.