This project has no private reporting channel. To report a security issue, open a GitHub issue titled "Security: " and say only that you found a vulnerability and which part it concerns (for example "the collector" or "the release workflow"). Please do not put exploit details, proof-of-concept code or anyone's personal data in the issue. The maintainer will reply there and arrange a way to receive the details.

Issues that are not sensitive (a hardening idea, a doc error, a missing check) can be reported in full in a normal issue or pull request.

This is a best-effort, volunteer-run project: there is no guaranteed response time. Fixes are released as a new version tag and noted in CHANGELOG.md.

Only the latest release receives fixes.

In scope: anything in this repository, in particular

• the scripts in semantic-core-expander/scripts/ (network access, file writes, parsing of Google responses and of project.json);
• the instructions in SKILL.md and references/ (for example, anything that could lead Claude to follow instructions found in collected data);
• the release pipeline (tools/build_zip.py, .github/workflows/).

Design guarantees the project aims to keep:

• the only network destination is https://suggestqueries.google.com/complete/search, and redirects are never followed;
• no credentials, telemetry, subprocesses, eval or exec;
• all writes stay inside the user's project directory;
• collected text is stripped of control and bidi characters before it is printed or saved;
• spreadsheet exports cannot contain live formulas from collected text;
• release zips contain only allowlisted files and pass a privacy scan, and the release job runs no third-party packages.

A change that breaks one of these is a security bug.

Out of scope: Google's availability or blocking behaviour, and the content of suggestions themselves (they are third-party public text).