Security

Tideline reads files that describe powerful access, so it is held to a strict standard. This page states what is enforced, what is not yet done, and how to report a problem.

Enforced by tests

• No network. A test replaces the socket layer and fails on any connection attempt. A source test forbids network and subprocess imports.
• Read-only. A test snapshots the scanned tree before and after a scan. A source test forbids write, delete and eval constructs.
• No secret values in output. A test plants fake credentials of several kinds and asserts none appear in text, JSON or Markdown output.
• Bounded reads. Files over 256 KB are skipped, directory depth is limited, symlinks are not followed.

Not yet done

• Human independent security review. Two AI-assisted audits were run and their findings fixed (docs/AUDIT_RESPONSE.md); the author's own review is SECURITY_REVIEW.md.
• Signed releases and published checksums.
• Trademark clearance for the name.

Limits of the tool

• The score is a heuristic for ranking, not a guarantee of safety.
• Detection relies on known file locations and patterns. Unknown tools and custom setups are not seen.
• It cannot see credentials held only in memory, keychains or remote services.

Report a vulnerability

Use GitHub private vulnerability reporting on the repository. Please do not open a public issue for a security problem. A contact address will be listed here once the project domain is live.