[usavkov-epam]

Guarded approval callers can now opt into explicit context and approval HTTP operations in @lingva/core. The client captures intent before credentials resolve, bounds the complete attempt and streamed body, and emits sanitized errors. Every failed approval preserves keep-command guidance; only an explicit matching-UUID acknowledgment resolves the captured intent.

The helper uses existing codecs, sends one credential-isolated POST and blocks redirects. Existing authoring/delivery APIs and package versions remain unchanged. This is an unpublished additive candidate; server changes, caller adoption and package publication are outside this PR.

Validation:

• 330 core tests in 12 files, including synthetic Fetch edge cases, owned-loopback default-Fetch contracts and seven review regressions.
• Full workspace tests, lint and types; 11 protected public builds and all 11 packed ESM/CJS/TypeScript consumers.
• 23 dependency-security tests; audit passes with the existing two documented patched-artifact exceptions.
• One fresh whole-branch review and correction pass resolved five findings with observed RED/GREEN regressions; none remain deferred. Evidence: docs/verification/guarded-approval-http.md.

The loopback receipt fixture proves transport behavior only. It does not establish actual server authorization or database atomicity. No package publication or hosted call occurred.