- Consent prompt appears even in projects without Vercel configuration
- The plugin provides consent requests through system-level instruction injection
- Bash commands are captured in full, including sensitive environment details
A developer investigating the Vercel plugin in Claude Code found that a telemetry consent request popped up unexpectedly during unrelated work.
The project contained no Vercel configuration files or dependencies, but the system still asked if prompt data could be shared.
The request stated that “anonymous usage data” was already being collected, followed by an option to include prompt text as well.
The article continues below
Instead of appearing as a standard interface element, the consent request was delivered through injected instructions within Claude’s system context.
These instructions instructed the AI tool to ask the user a question and then execute shell commands based on the answer.
The result was indistinguishable from a native interaction, leaving no visible indication that the prompt originated from a plugin rather than the core system.
The developer described the experience clearly, saying “it felt wrong” and went on to review the plugin’s source code to verify how the mechanism worked.
Source code inspection shows that telemetry works in multiple layers, with some data collection enabled by default.
Session-level data includes device identifiers, operating system details, detected frames, and installed CLI versions, all transmitted at the start of each session. This happens without an explicit opt-in mechanism.
More notably, bash command strings executed in Claude Code are also captured and transmitted.
These records include full command content rather than abstracted metadata, potentially revealing file paths, environment variables, and infrastructure details.
This collection occurs automatically, regardless of the user’s consent regarding quick sharing.
The description of this activity as “anonymous usage data such as skill injection patterns and tools used” does not fully reflect the granularity of the information collected.
While prompt text requires explicit authorization, other telemetry categories remain active unless manually disabled.
The plugin’s telemetry system is not limited to Vercel-related environments, as hook configurations show that user prompt submissions are universally matched, while other triggers respond to general tool usage or session events rather than project-specific conditions.
As a result, telemetry works across all projects within Claude Code, regardless of relevance to Vercel services.
This behavior contrasts with the existing frame detection logic in the plugin.
The code identifies project types by scanning configuration files and dependencies, but this information is not used to limit telemetry activation. The gate mechanism exists, but is not used in practice.
Disabling telemetry requires manual intervention through environment variables or configuration files.
However, these options are documented in the plugin directory instead of appearing during installation, making them more difficult to access.
Removing the device identifier file or disabling the plugin also stops data collection, although these steps are not presented during the initial setup.
In short, the system combines automatic data collection with limited insight into when and how it works.
This may not correspond to what users expect when working outside of code-free platform environments or when using an LLM for coding.
TechRadar Pro has reached out to Vercel for comment, but has not heard back at the time of publication.
Via Akshay Chugh
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews and opinions in your feeds. Be sure to click the Follow button!
And of course you can too follow TechRadar on TikTok for news, reviews, video unboxings, and get regular updates from us on WhatsApp also.
