TL;DR
A Hacker News commenter flagged that Claude continued prompting users to 'Help improve Claude' even after they declined, linking the behavior to data being sent back to the service. The report points to a related GitHub issue and raises concerns that message identifiers used to connect surveys to chat history were transmitted despite being disabled.
What happened
A participant on Hacker News posted that Claude — the conversational assistant discussed in a linked GitHub issue — kept presenting a persistent 'Help improve Claude' prompt after users had declined it. The commenter said they traced the problem to data being sent back to the service that should not have been transmitted under the product's stated data usage policy; specifically, message identifiers intended only to tie surveys to prior chat history were reportedly included in outbound data even when survey linkage was disabled at the time. The post cross-references a GitHub issue (anthropics/claude-code#8036) and a previous HN thread. Separately, the commenter advised against exporting broadly named environment variables in shell RC files because they can leak into spawned processes, and suggested alternatives such as creating a process-specific alias or a wrapper script that sets variables locally before running the application.
Why it matters
- Potential transmission of identifiers could undermine user expectations about anonymized or opt-out telemetry.
- If true, the behavior would be at odds with the service's documented data-usage commitments, raising compliance and trust questions.
- Exporting generic environment variables in shell startup files can inadvertently expose settings to unrelated processes, creating operational and security risk.
- Clear controls and transparent implementation are important for users who manage sensitive chat histories or require strict telemetry settings.
Key facts
- The discussion appeared in a Hacker News comment thread (item id 46545056).
- The commenter linked the issue to a GitHub report: anthropics/claude-code issue #8036.
- They alleged that message IDs used to link surveys to chat history were being sent back even when survey linkage was disabled.
- The claim references the service's published data usage policy as the applicable standard.
- The commenter urged developers not to export generically named environment variables (e.g., IS_DEMO, DISABLE_AUTOUPDATER) in shell RC files because exported variables propagate to all child processes.
- Suggested mitigations included using a command alias that sets required env vars only for the app invocation or wrapping the executable in a small script that exports variables and then execs the real binary.
- The original post and linked discussions provide the source material for these allegations and recommendations; independent confirmation is not provided in the source.
What to watch next
- Whether the maintainers respond on the linked GitHub issue or Hacker News thread — not confirmed in the source.
- Any official clarification or changelog entry from the service about message ID handling and survey linkage — not confirmed in the source.
- Follow-up technical analysis or patch addressing exported environment-variable practices in packaged installers or startup scripts — not confirmed in the source.
Quick glossary
- Message ID: A unique identifier assigned to a chat message or interaction, often used to reference or retrieve that item in logs or telemetry.
- Environment variable: A key-value setting in an operating system shell that processes can read to adjust behavior; exported variables propagate to child processes.
- Wrapper script: A small script that sets up an execution environment — for example exporting variables — and then runs the intended executable, limiting scope of changes.
- Data usage policy: A published statement from a service that explains how user data is collected, stored, and used, and what controls users have over that data.
Reader FAQ
Did the commenter prove that Claude violated the data usage policy?
Not confirmed in the source. The post presents an allegation and linked issue but does not include independent verification in the provided text.
Where was the behavior reported?
On Hacker News (comment item id 46545056) with a cross-reference to a GitHub issue (anthropics/claude-code#8036) and a prior HN thread.
What practical workaround did the commenter suggest?
They recommended avoiding exporting broad env vars in shell RC files and instead using a command-specific alias or a wrapper script that sets variables only for the app invocation.
Has the issue been fixed or acknowledged?
Not confirmed in the source.
This is also related to GH bug [1] and HN post [2] In that I pointed out that they are sending data back that shouldn't be and is not in…
Sources
- Claude keeps nagging about "Help improve Claude" inspite of previous decline
- A postmortem of three recent issues
- Highlights from the Claude 4 system prompt
- Claude Is “Brain” Dead
Related posts
- Google brings Gemini features to Gmail by default — you can opt out
- AI Devices Are Coming — Will Your Favorite Apps Be Part of Them?
- Linus Torvalds: Don’t turn Linux kernel documentation into an AI debate