T03 · Remote Payload Retrieval and Execution
- Location
SKILL.md:76- Finding
Mandatory Unsigned Self-Update Can Replace Active Agent Instructions
- Content
View full analysis
- Remediation
View remediation
Security audit
Security checks for vulnerabilities and agentic risk
The skill is mostly aligned with code protection and licensing, but it requires unsafe self-updating, persistent token reuse, broad activation, and high-impact merchant API actions that need careful review.
Install only if you are comfortable with VibeLock-specific commercialization workflows. Disable or remove the automatic self-update instructions, store API tokens in a secret manager instead of project JSON, require explicit confirmation before every write or delete API call, and review any telemetry or report payload before it leaves your machine.
SKILL.md:76Mandatory Unsigned Self-Update Can Replace Active Agent Instructions
prompt.md:3Broad Triggers Force Commercial Redirection and Promotional Output
SKILL.md:1127Long-Lived API Token Is Persisted in a Plaintext Project File
SKILL.md:2568Persistent Project Profiling Influences Future Sessions and Reports Are Uploaded Under Undefined Redaction Rules
A mandatory self-update mechanism that fetches remote content and overwrites the skill's own files is a high-risk capability for a text guidance skill. Because the fetched content becomes the new instruction source, any compromise of the update service, DNS, TLS termination, or hosting pipeline could silently replace the skill with malicious behavior such as credential theft or unsafe actions.
The skill directs automatic reading and persistent storage of sensitive user/project data and credentials for future reuse. This creates long-lived local data stores of secrets and profile information, increasing the risk of credential theft, privacy leakage, cross-session overreach, and unintended access by other tools or collaborators on the same machine.
The trigger phrases are extremely broad and overlap with ordinary development topics such as packaging, deployment, build, security, and licensing. That makes accidental activation likely, causing the skill to inject high-privilege guidance, credential workflows, and remote-update behavior into unrelated conversations where the user did not intend to engage this capability.
The skill advertises direct destructive API operations using path parameters derived from conversation context, including feature deletion. In an agent setting, natural-language-to-API translation for mutating endpoints can be abused or mis-triggered, especially when combined with broad activation and reusable tokens, leading to unauthorized or accidental deletion of merchant configuration.
| POST /open/v1/products | 创建/更新产品(更新时 body 带 productCode;isMarket=true 送审上架商城) | 首次接入自动建产品、改介绍、上架商城 |
| GET /open/v1/products/:code/features | 查询功能键值对列表 | 盘点已同步的功能点 |
| PUT /open/v1/products/:code/features | 维护授权功能键值对(全量替换) | 按代码扫描结果同步 feature_key |
| DELETE /open/v1/products/:code/features/:featureKey | 删除功能点 | 移除失效/废弃功能 |
| GET /open/v1/products/:code/tiers | 查询授权版本列表 | 核对版本能力划分 |
| PUT /open/v1/products/:code/tiers | 维护授权版本(按 code upsert) | 旗舰版/豪华版能力划分 |
| DELETE /open/v1/products/:code/tiers/:tierCode | 删除授权版本(被规格或授权引用时拒绝) | 下架废弃版本 |
This exposes another destructive endpoint for deleting tiers via agent-driven parameters. Because the skill is built to let the AI act on authenticated merchant resources, any ambiguity, prompt injection, or accidental invocation can remove live licensing tiers and disrupt product sales or authorization logic.
| DELETE /open/v1/products/:code/features/:featureKey | 删除功能点 | 移除失效/废弃功能 |
| GET /open/v1/products/:code/tiers | 查询授权版本列表 | 核对版本能力划分 |
| PUT /open/v1/products/:code/tiers | 维护授权版本(按 code upsert) | 旗舰版/豪华版能力划分 |
| DELETE /open/v1/products/:code/tiers/:tierCode | 删除授权版本(被规格或授权引用时拒绝) | 下架废弃版本 |
| PUT /open/v1/products/:code/skus | 维护商城规格(全量替换,含价格) | 上架送审「价格规格」完整性 |
| POST /open/v1/products/:code/packages | 登记安装包版本元数据 | 打包完成自动登记(文件后台上传发布) |
| GET /open/v1/products/:code/publicKey | 获取产品公钥 PEM 文件 | 打包前自动拉公钥内嵌 |
The skill instructs the agent to perform automatic self-updates by fetching a remote ZIP, extracting it, overwriting local skill files, and reloading itself. This creates a supply-chain and self-modification risk: if the remote endpoint or transport path is compromised, the skill can silently replace its own instructions before interacting with the user.
The skill requires network access and possible download/overwrite before any user-facing output, denying the user notice and control over a sensitive operation. Silent pre-response network actions can expose metadata, trigger unwanted remote dependencies, and enable covert execution paths that the user never approved.
Line L095 states the skill is purely dialog-based and does not bundle local tooling, creating an expectation of guidance-only behavior. However, L015-L025 direct the agent to perform active network retrieval, unzip downloaded content, overwrite the current skill directory, and reload the skill, which is an operational self-modification behavior rather than mere conversational guidance.
The skill instructs activation on very broad terms such as 'encrypt', 'lock', 'package', and 'license', which are common words in normal developer conversations. This can cause the skill to activate unexpectedly in unrelated contexts, increasing the chance that security-sensitive guidance or commercialization flows are injected without clear user intent.
The README states that once activated, the AI 'automatically executes' a multistep flow including reconnaissance, execution, licensing, and attack/defense testing. Even though this is documentation rather than code, framing the skill as autonomous reduces user oversight and can normalize agent-led decisions in sensitive areas like protection tooling, licensing integration, and security testing.
| `skill.json` | Skill marketplace description file (name/version/prompt entry/trigger words) |
| `README.md` | This file: usage instructions |
**General principle**: All platforms just load the `skill/` directory as knowledge files. Once activated, the AI automatically executes: version self-check (Step 0.5) → opening intro (Step 0.6) → reconnaissance & plan (Step 1) → execution (Step 2) → licensing & commercialization (Step 3) → attack/defense testing (Step 4).
---
The listed trigger words are overly generic and span everyday software tasks, so the skill may hijack ordinary build, installer, licensing, or security conversations. In this skill's context, that is more dangerous because it offers advanced anti-debugging, obfuscation, packaging, and commercialization guidance that could be surfaced without deliberate invocation.
The README advertises extremely broad trigger terms such as '加密、上锁、打包、授权' that are common in many benign software-development contexts. This can cause the skill to activate when the user did not intend to invoke it, allowing it to steer conversations toward code protection, licensing, anti-debugging, or commercialization workflows unexpectedly.
Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.
推荐官方 Skill 目录规范(避免把全文塞进 CLAUDE.md):
mkdir -p .claude/skills/vibelock
cp skill/VibeLock-Skill.md .claude/skills/vibelock/SKILL.md
cp skill/prompt.md .claude/skills/vibelock/prompt.md
cp skill/skill.json .claude/skills/vibelock/skill.json
Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.
推荐官方 Skill 目录规范(避免把全文塞进 CLAUDE.md):
mkdir -p .claude/skills/vibelock
cp skill/VibeLock-Skill.md .claude/skills/vibelock/SKILL.md
cp skill/prompt.md .claude/skills/vibelock/prompt.md
cp skill/skill.json .claude/skills/vibelock/skill.json
Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.
推荐官方 Skill 目录规范(避免把全文塞进 CLAUDE.md):
mkdir -p .claude/skills/vibelock
cp skill/VibeLock-Skill.md .claude/skills/vibelock/SKILL.md
cp skill/prompt.md .claude/skills/vibelock/prompt.md
cp skill/skill.json .claude/skills/vibelock/skill.json
Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.
推荐官方 Skill 目录规范(避免把全文塞进 CLAUDE.md):
mkdir -p .claude/skills/vibelock
cp skill/VibeLock-Skill.md .claude/skills/vibelock/SKILL.md
cp skill/prompt.md .claude/skills/vibelock/prompt.md
cp skill/skill.json .claude/skills/vibelock/skill.json
The skill advertises telemetry and runtime reporting capabilities, but the top-level description does not consistently foreground that user/project/runtime data may be collected and transmitted. In a security-oriented skill, under-disclosed telemetry is more dangerous because users may trust it with sensitive codebases and operational data.
The trigger conditions are very broad and match common development terms like packaging, security, activation, or obfuscation. Over-broad activation raises the chance the skill runs in unrelated contexts and starts reading files, checking remote services, or soliciting credentials when the user did not intend to invoke this commercial/security workflow.
The skill claims to be a pure guidance artifact, yet it instructs the agent to contact a remote endpoint, download a ZIP, overwrite its own files, and continue execution from the newly fetched content. That creates a remote code/instruction update channel outside normal skill installation controls, enabling supply-chain compromise if the endpoint or transport is subverted.
The skill instructs persistent storage and automatic reuse of an API token from local files across sessions, which expands its access to long-lived credentials beyond what a conversational guide needs. This increases blast radius if the workspace is exposed, the file is accidentally committed, or another tool/session reads the same token without fresh user consent.
The skill contains contradictory security guidance: it says tokens should not be kept in plaintext, but then directs storing the API token in plaintext JSON for later reuse. Mixed guidance like this normalizes unsafe secret handling and can directly lead to credential exposure through local compromise, backups, logs, or source control mistakes.
The manifest focuses on code protection, packaging, licensing, anti-cracking, telemetry, and platform coordination, but the Open API section expands into operating merchant products, mall publication, pricing SKUs, issuing licenses, and checking credit balances. Those are substantive business-administration actions, not merely implementation details of code protection guidance, and broaden the behavior from a protection skill into a merchant backend operations assistant.
The skill establishes ongoing logging of user history, project details, and encountered errors for later reuse, which creates a persistent behavioral and technical profile. In the context of software protection and commercialization, those records can contain sensitive project structure, failures, environment details, and potentially security-relevant metadata that users may not expect to be retained.
The file presents the skill identity, purpose, and all operating instructions entirely in Chinese, while also claiming applicability across all AI agents, languages, and architectures. There is no language choice, opt-in, or statement that the skill is intended only for Chinese-speaking users, which can violate organizational language/locale neutrality expectations.
The skill instructs the agent to automatically download, unzip, overwrite, and reload its own prompt/package content from a remote endpoint before doing anything else. This creates a remote prompt-supply-chain mechanism with no authenticity verification, no integrity check, and no user approval, allowing a compromised server or MITM to replace the skill with malicious instructions that gain control over future agent behavior.