#377 test(claude): prove unrecoverable delete parity

[Anonymous]

Originally created by: jasperdingg

Pull Request

Slice

• Repo: doberman-core
• Feature / Slice: [#329]
• Plan reference: [#329]

What this PR does

Marks the existing Claude Code test as proof that deleting unrecoverable gitignored data requires AUTH, and updates the parity matrix.

Tests added (run in CI)

• Focused tests, full suite, parity, Ruff, formatting, and import checks pass.
• Mutation check confirms the test fails if the delete safeguard is removed.

Public-release safety (doberman-core only)

• [x] Contains nothing from the "not allowed" list: no enterprise/hosted code, no proprietary detection, no customer data, no secrets, no commercial-license code
• [x] Core still builds/tests/runs with NO enterprise package installed

Security checklist

• [x] Fails closed on error / uncertainty
• [x] No secret, full file, or unredacted prompt logged or committed
• [x] Any guardrail/learning change is raise-only (no silent loosening)
• [x] Every BLOCK/AUTH carries reason codes + a human explanation
• [x] doberman-core does not import doberman_enterprise

Edge cases covered / Deviations from plan / Risks introduced

• rm data/app.db requires AUTH.
• The operand is not exposed in the reason.
• No production behavior changed.

[Anonymous]

Ticket changed by: fu351

• status: open --> closed

[Anonymous]

Originally posted by: fu351

Merged, and this closes [#329]. The AN-1 gate now has its Claude Code host-hook proof to sit beside the engine-level one.

What stood out: you didn't stop at proving the verdict. assert "data/app.db" not in reason proves the redaction invariant that makes the guarantee safe to ship — the AUTH message can't become an oracle for which unrecoverable file exists at a given path. That's the assertion most people skip, and it's the one that matters here.

I ran the mutation check before merging (neutralizing _rm_targets_unrecoverable_data to a no-op sends the test red, exit 1 — reverted). And you let the generator produce the PARITY.md cell rather than hand-editing it, so CI's own test_parity_matrix.py --check confirms the doc matches the marker. Clean, well-scoped work.

Next rung on the parity ladder if you want it: [#330] is the same AN-1 cell on the MCP proxy — you've now done this exact shape end-to-end, so it should go quickly.