[BenWestgate]

Use the injected entropy boundary from #200 to source every fresh CLI draw from CoreEntropy: hash a fresh Bitcoin Core xprv with 32 independent OS-random bytes, and fail closed if Core or temporary-wallet cleanup fails.

The temporary wallet is encrypted, relocked and unloaded for each draw. Supported termination signals become cleanup interruptions before another card is emitted. Security/user docs describe RAM-backed storage, disk fallback and leaked-wallet limits.

Validation: 347 Core/CLI/generation tests pass on the rebased split; git diff --check passes. The implementation was independently security-reviewed before the split, including a 22/22 Linux/Core 32.0rc4 run; the new head still needs a current-head review after this rebase.

The standalone verification harness and CI are split into a separate stacked PR.

[BenWestgate]

@codex security-review

---

Generated by Claude Code

[chatgpt-codex-connector]

Summary

• Completed the security review and fixed a fail-open cleanup condition: failures from Bitcoin Core’s unloadwallet RPC are no longer silently ignored. Filesystem and RAM-disk cleanup is still attempted, after which the operator receives actionable guidance to stop Bitcoin Core. src/codex32/_bitcoin_core.pyL516-L535
• Preserved the cleanup state when the temporary wallet directory cannot be removed, allowing a subsequent cleanup attempt rather than prematurely treating the entropy wallet as closed. src/codex32/_bitcoin_core.pyL528-L531
• Updated the regression test to verify that an unload failure is surfaced while the temporary wallet directory is still removed on a best-effort basis. tests/test_bitcoin_core.pyL909-L925
• Committed the changes as d47ef3e (Report entropy wallet unload failures) and created the follow-up pull request metadata with title “security: Surface entropy wallet unload failures.”

Testing

• ✅ PYTHONPATH=src python -m pytest -q tests/test_bitcoin_core.py — 52 passed.
• ✅ python -m ruff check .
• ✅ python -m ruff format --check .
• ✅ PYTHONPATH=src python -m mypy src/codex32
• ✅ git diff --check
• ⚠️ python -m pip install -e '.[dev]' — package-index access was blocked by the environment with HTTP 403, so missing development dependencies could not be installed.
• ⚠️ PYTHONPATH=src python -m pytest -q tests/test_cli.py — could not run correctly without installed codex32 distribution metadata after the blocked installation.
• ✅ git status --short --branch — working tree is clean on branch work.

View task →

[BenWestgate]

@codex we don't see the changes you speak of.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	⚠️ Failed 2026-10-24T11:17:20.039678Z We couldn’t complete this review. Comment @codex review to try again.	0d5df21	New commits

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[BenWestgate]

cACK # Hashing keeps the plaintext xpub in the wallet file from checking guesses of a share.

The question then becomes, how much entropy is in an xprv and why SHA256 instead of another hash function for example SHA512? And why just one round?

[chatgpt-codex-connector]

Summary

• Implemented the previously described security changes so a failed Bitcoin Core unloadwallet RPC now raises an actionable error after best-effort filesystem and RAM-disk cleanup, rather than being reduced to a warning. src/codex32/_bitcoin_core.pyL516-L534
• Preserved the temporary wallet and RAM-disk state when directory removal fails, allowing a subsequent close() call to retry cleanup. src/codex32/_bitcoin_core.pyL525-L530
• Added regression coverage verifying that unload failures are surfaced even when the temporary files are removed, and that failed directory cleanup can be retried successfully. tests/test_bitcoin_core.pyL910-L946
• Committed the changes as ad57890 (Report entropy wallet unload failures) and created the follow-up PR metadata titled “security: Surface entropy wallet unload failures.”

Testing

• ✅ PYTHONPATH=src python -m pytest -q tests/test_bitcoin_core.py — 53 passed.
• ✅ PYTHONPATH=src python -O -m pytest -q tests/test_bitcoin_core.py — 53 passed, with the expected pytest warning about disabled assertions under optimized Python.
• ✅ python -m ruff check .
• ✅ python -m ruff format --check .
• ✅ PYTHONPATH=src python -m mypy src/codex32
• ✅ git diff --check
• ⚠️ PYTHONPATH=src python -m pytest -q && PYTHONPATH=src python -O -m pytest -q — full-suite collection could not complete because the environment does not have the hypothesis dependency installed.
• ✅ git status --short --branch — working tree is clean on branch work.

View task →

[BenWestgate]

Security review of 40fb74f (PR #145; 13 changed files).

[P2] Treat failure to unload the entropy wallet as a creation failure. In CoreEntropy.close(), unloadwallet errors are reduced to a warning, _folder/_disk are cleared before cleanup succeeds, and rmtree(..., ignore_errors=True) hides filesystem removal failures. _create() can consequently report a successful ceremony (or proceed to wallet initialization) while the entropy wallet remains loaded in Bitcoin Core; callers also lose the state needed to retry cleanup. The source xprvs held in that wallet determine all the generated entropy. Although the wallet is normally relocked after each draw, combined failures of walletlock and unloadwallet can leave the keys RPC-accessible during the 60-second unlock window. Propagate unload failure after best-effort cleanup, preserve failed resource handles for retries, and test this at the CLI boundary. The current unit test explicitly expects a warning and successful close() on unload failure.

Other security boundaries reviewed: Bitcoin Core selection and stdin-only passphrase handling; 256-bit passphrase generation; SHA-256 extraction from each fresh xprv to prevent stored public xpubs from testing guesses of shares; explicit source injection and byte-count checks; unbiased index rejection sampling; confirmation-gated shares; OS-specific RAM/disk paths and signal cleanup. SHA-256 is appropriate here: one invocation already provides 256-bit preimage resistance, and SHA-512 or repeated hashing cannot increase entropy beyond the source. No other actionable vulnerability was identified in these paths.

CI: 10 Python test-matrix jobs and 1 Bitcoin Core fixture job passed. The managed local diff scanner could not start because head 40fb74f is not available in the local Git object database; this review inspected GitHub's exact PR patches and head-revision source. Recommendation: revise before merge and re-review wallet teardown.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: d20d70dd00

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[BenWestgate]

AI-assisted current-head security/code review.

Concept ACK bea2914; not a security ACK yet. Seed, share, index and identifier draws now require injected entropy; each SHA-256 draw combines fresh Core xprv material with independent OS bytes, with no silent fallback on Core failure. Wallet cleanup and interruption handling have tests. Before human approval: independent real-Core 32.0rc3 verification of repeated draws, interruption/unload, disk-fallback warning and secret-file removal, plus security review of the new CoreEntropy boundary.

[BenWestgate]

This is the real-Core check the review above asks for, run on bea2914 against Bitcoin Core 32.0rc3 regtest on Linux. All 25 checks passed.

• Repeated draws: 6 draws of 1 to 64 bytes. Each one used a new encrypted wallet in /dev/shm. Each wallet was locked before unload, then unloaded and its folder deleted before the bytes came back. The 64-byte draw took two keys, and all draws differ.
• Interruption: a real SIGINT landed while each RPC was in flight (createwallet, walletpassphrase, addhdkey, gethdkeys, walletlock, unloadwallet). Each time, no wallet stayed loaded and no folder was left. A SIGINT before Core ran unloadwallet stopped creation with "Restart Core to unload it". That wallet's deleted folder did not come back when Core later unloaded it.
• Real CLI in a terminal: killed mid-draw by SIGTERM, SIGHUP and SIGINT. Each run exited 130 with the void notice and left nothing.
• Disk fallback (no /dev/shm): one warning, naming the folder and the TMPDIR fix. Each wallet folder was deleted from disk before the bytes came back.
• Secrets: Core's debug.log holds no tprv and no passphrase, and Core's wallets folder holds no temporary wallet.

Not covered: the macOS RAM disk and Windows against real Core. CI covers those paths only with fakes. I wrote both the change and this check, so it isn't independent. The script is about 250 lines and could go under tools/ if you want it rerunnable here.

---

Generated by Claude Code

[BenWestgate]

cACK: The script is about 250 lines and could go under tools/ if you want it rerunnable here.

Yes.

[BenWestgate]

Do the commits follow contributing.md?

[BenWestgate]

The 14-commit history contained repeated fixups and reversals, so it needed consolidation under CONTRIBUTING.md's squashing guidance. It is now one atomic commit, 7cb2ae6, with a short component-prefixed subject and a body explaining the final behavior, security implications and validation.

Verified git diff bea2914 7cb2ae6 is empty and both commits have tree a733bd982033c49f6c8cfb1a020bde376bc29a81. No file changed. The reviewed code and existing passing checks therefore apply to identical contents; fresh CI is running for the new SHA. PRs #159 and #160 were restacked in the same atomic push, with their own complete trees also unchanged.

All six inline threads are resolved. The approved standalone verification script is still absent from the branch, so the PR description now explicitly tracks that request and independent security review as remaining work. This history cleanup does not supply either.

[BenWestgate]

AI-assisted incremental review at 753eded: Code ACK for the follow-up test/CI changes. The standalone 22-check Core entropy harness is committed, the workflow installs codex32 before calling its CLI, and both current-head CI jobs passed. Independent security review of CoreEntropy and real-Core macOS/Windows coverage remain outstanding. Not a security/merge ACK.

[BenWestgate]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Another round soon, please!

Reviewed commit: 753eded110

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[BenWestgate]

AI-assisted independent security review at 753eded: 22/22 entropy harness checks passed on real Bitcoin Core 32.0rc4 (Linux regtest), including fresh draws, per-draw wallet cleanup, interruptions, disk fallback and log checks. No confirmed vulnerability in the 15-file diff. Linux Core boundary verified; full Tails creation/restore and macOS/Windows runtime coverage remain open. Not yet a merge ACK.

[BenWestgate]

Not going to code review +842

Please avoid super pull requests which attempt to do too much, are overly large, or overly complex as this makes review difficult.

I cACK the description and most of the approach but you need to make this easier to review somehow.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[BenWestgate]

Superseded by focused review units #200 (entropy-source API), #201 (Core/CLI integration), and #202 (standalone Core verification). GitHub would not retarget this stacked PR, so keeping #145 open would duplicate their combined diff.