[ericliswan]

Summary

Implements pwv init. It prompts for the master password twice, rejects passwords shorter than 12 characters or with a zxcvbn score below 3, and creates an empty vault. It refuses to replace an existing vault unless --force is given; in that case the old file is kept as <vault>.bak.

Closes #28

Type of change

• feat - [ ] fix - [ ] docs - [x] test - [ ] refactor - [ ] ci/chore - [ ] security

How was this tested?

• cargo fmt --all --check, cargo clippy --workspace --all-targets --all-features -- -D warnings (and without features), cargo test --workspace --all-features, cargo deny check
• Unit tests for the strength check: too short, weak (error never contains the password), strong
• Integration tests through the real binary: the new vault unlocks with the master password; a second init fails with exit 1 and leaves the file unchanged; --force replaces it and keeps the old bytes in .bak; mismatched and weak passwords fail without creating a file or echoing the password
• By hand: init, re-init, --force and weak passwords on macOS

Security checklist

• No secret in args, env vars, logs, Debug output, errors or panics (strength errors repeat zxcvbn's fixed feedback text only)
• Crypto is called only from pwv-core::kdf / pwv-core::crypto
• Any new header/file-format field is inside the AAD and validated on load (no format change)
• Failure paths have tests (wrong password, tampering, bad input)
• Docs (README / DESIGN / THREAT_MODEL) updated if behaviour changed (no doc change needed until the v0.5 README)

Notes for the reviewer

• The strength check runs before the confirmation prompt, so a weak password fails on the first entry.
• zxcvbn keeps pieces of the password in its result without zeroizing them. The result is dropped straight away; this goes in the README limitations about memory copies.
• New feature pwv-cli/insecure-test-kdf forwards to pwv-core and is off by default. Without it, the vault-creating CLI tests are ignored with a reason, because the production KDF in a debug build took 23 s for three tests. CI's --all-features run executes them.
• New dependency: zxcvbn 3.1.1 (MIT). It pulls in derive_builder, fancy-regex, itertools and lazy_static, and cargo deny is clean.

[ericliswan]

Self-review: read the full gh pr diff. Checked the following:

• Secrets are exposed only for the length count, zxcvbn, and the confirmation comparison; nothing derived from them is printed.
• The strength error is built from zxcvbn's fixed Warning/Suggestion text, and the tests assert the password doesn't appear in it.
• Overwrite protection checks try_exists, so a permission error surfaces instead of being read as "no vault".
• --force relies on storage's existing .bak copy, and an integration test checks that.
• The release build has no insecure-test-kdf; the feature is opt-in only.
• 180 changed lines excluding Cargo.lock.

Not covered: interactive hidden prompts on a real TTY (manual only).