tests: BIP-39 mnemonic detector false-positive corpus #135

You will build a false-positive corpus that locks in the README's claim: English prose that uses wallet words never trips the mnemonic detector.

How it fits together

flowchart TD
    A["tests/test_mnemonic.py<br/>new FP corpus (near line 56)"] --> B["English sentences with 12+<br/>consecutive BIP-39 words"]
    F["bip39_words.py<br/>WORDS list"] --> B
    B --> C["scan_text() -> detect_mnemonics()<br/>(mnemonic.py line 58)"]
    C --> D["_bip39_checksum_ok() (line 35)<br/>rejects the window"]
    D --> E["assert no bip39-mnemonic finding"]

Wordlist membership alone would match prose, so detect_mnemonics() confirms a real BIP-39 checksum (or an Electrum HMAC tag) before it reports anything. These tests pin that guarantee from the false-positive side; the existing VALID_12 / VALID_24 cases already pin the true-positive side.

Steps

1. In tests/test_mnemonic.py, build a small corpus of English sentences that contain 12 or more consecutive words drawn from the BIP-39 list (src/agentsweep/bip39_words.py, which is full of common words like "abandon", "expose", "actor") but do not form a valid checksummed mnemonic.
2. Assert detect_mnemonics returns no bip39-mnemonic finding for every sentence. Drive it through scan_text, the way the other tests do.
3. Place the corpus next to the existing positive cases so both directions of the claim stay covered.

Files

• tests/test_mnemonic.py (extend near test_natural_english_not_detected, line 56)
• Code under test: src/agentsweep/mnemonic.py (detect_mnemonics, line 58)

How to claim

Comment "I'll take this", fork the repo, branch from main, and open a PR that links this issue.